Research Status and Prospect of Target Tracking Technologies via Underwater Sensor Networks
-
摘要: 水下目标跟踪在海洋资源的开发利用以及国家安全的防御等方面都具有广泛的应用价值和重要的战略意义. 基于水下传感器网络(Underwater sensor networks, USNs)的目标跟踪技术凭借其覆盖范围广、观测时间长和实时融合等优势已经成为一个新的研究热点. 本文针对基于USNs的目标跟踪关键技术的基本思想、研究进展、应用及局限性进行了综述, 主要从以下几个角度对其展开论述: USNs的建设现状、系统组成及其分类、目标跟踪系统模型、单目标跟踪技术、多目标跟踪技术以及能效优化措施. 最后, 本文不仅指出了基于USNs的目标跟踪研究目前存在的主要挑战, 并对该领域的未来发展方向进行了展望.Abstract: Underwater target tracking has extensive application value and important strategic significance in the development and utilization of marine resources, and national security defense. Target tracking technology via underwater sensor networks (USNs) has become a new research hotspot for its wide coverage, long observation time and real-time fusion. This paper gives a review of the basic ideas, research progresses, applications, and limitations of those key technologies of target tracking via USNs. It is mainly discussed from the following respects: the construction status of USNs, system composition and classification, target tracking system model, single target tracking technoligies, multi-target tracking technoligies, and energy efficiency optimization measures. Furthermore, this paper not only points out the main challenges of researches of target tracking via USNs, but also prospects the future development direction of the field.
-
现代过程工业中存在着一类间歇过程[1], 如半导体加工、制药、注塑、发酵等.间歇过程通常具有重复特性, 且对跟踪精度要求较高, 是典型的非连续操作.其控制任务是在每个生产批次内跟踪给定的参考轨迹[2].迭代学习控制(Iterative learning control, ILC)能够利用过去批次的信息进行优化学习, 不断调整控制输入轨迹, 逐步提高跟踪性能, 实现对参考轨迹的高精度跟踪, 因此被广泛应用于间歇过程控制中[3].但是由于ILC是典型的开环控制, 因此不能保证控制系统的时域稳定性, 难以处理实时干扰.模型预测控制(Model predictive control, MPC)作为先进过程控制技术[4], 不仅广泛应用于工业过程的优化控制[5-6], 同时也成功应用于轨迹跟踪控制[7].它通过预测未来的系统状态及输出, 进行滚动时域优化, 能及时处理实时干扰, 保证时域跟踪性能及闭环系统稳定性, 弥补ILC的不足.迭代学习模型预测控制(Model predictive iterative learning control, MPILC)结合了MPC与ILC的优点, 因此成为控制间歇过程的有效方法.
早期的MPILC算法大多基于输入输出模型, 如受控自回归积分滑动平均(CARIMA)模型[8-9], 脉冲响应模型[10].而近年来基于状态空间模型的MPILC算法研究受到了更多关注, 研究对象包括线性定常状态空间模型[11]、带干扰项的状态空间模型[12]以及含不确定性的状态空间模型[13].其控制器设计通常需要进行状态增广以构造二维误差模型.为加强控制器鲁棒性, 很多学者在此基础上提出控制器结构改进算法, 如构造分段优化[14], 改善学习机制[15].
典型的间歇过程通常具有强非线性, 而现有的MPILC算法大都是针对线性系统构造的.文献[14$-$15]将原非线性系统在工作点简单线性化, 文献[10]沿参考轨迹进行线性化.由线性化带来的模型失配问题会在一定程度上影响时域跟踪性能以及迭代学习速度, 这在实际生产上将造成原料浪费以及经济效率下降.近年来许多学者在MPILC研究中通过各种建模手段来近似非线性系统, 包括T-S模糊建模[16]、神经网络建模[17]、数据驱动建模[18]等.但是这些方法需要大量过程数据的支持, 以及极其复杂的调参、学习过程才能建立较为精确的模型.
线性参变(Linear parameter varying, LPV)蕴含技术是处理复杂非线性的有效手段, 已被广泛应用于非线性模型预测中[19].原非线性系统在工作区间的动态特性可以包含在由LPV系统构成的多胞里.因此, 只要保证基于LPV模型的控制系统的稳定, 就能够保证非线性控制系统的稳定[20].由于LPV模型中存在的参数不确定性, 其控制求解一般通过线性矩阵不等式(Linear matrix inequality, LMI)约束下的目标函数优化来实现.
间歇过程的参考轨迹会由于不同的产品规格、生产效率以及外在干扰而发生改变.比如半导体制造中的蚀刻系统必须跟踪不同的操作轨迹来生产不同规格的晶片[21].而一旦参考轨迹发生变化, 经典迭代预测控制需要重新进行初始化, 并经历多个批次的学习来跟踪新轨迹[10], 适应能力较差.近年来许多学者提出改进的自适应ILC算法来解决变参考轨迹跟踪控制问题.文献[21]针对随机系统变轨迹跟踪问题, 提出两种自适应ILC策略, 一是在控制器设计中选择当前批次的输出轨迹与下一批次的参考轨迹的差值作为新的状态变量, 另一种是在每一个批次的末尾利用卡尔曼滤波器重新进行系统辨识; 文献[22]针对离散非线性系统, 设计模糊自适应ILC控制器, 通过不断更新模糊参数来近似变参考轨迹下的系统动态; 文献[23]设计基于数据驱动的自适应ILC控制器, 引入未来批次的参考轨迹作为反馈, 并将过去批次的参考轨迹作为前馈以实现对变参考轨迹的跟踪.这些基于ILC的控制策略通常要求已知所有采样时刻上的参考轨迹变化量, 并且由于算法内缺少预测环节导致了跟踪性能和学习效率的下降.针对该问题, 可将参考轨迹变化量视为迭代域中存在的有界扰动, 构造限制参考轨迹变化量对系统控制性能影响的$H_\infty$约束.这样只要在每个采样时刻的优化中满足此$H_\infty$约束, 就可有效抑制变参考轨迹带来的跟踪误差波动, 且只需已知下一采样时刻的参考轨迹变化量. $H_\infty$控制[24]可与MPILC算法有效结合, 利用预测控制的滚动时域优化提高时域跟踪性能, 从而加快学习速度.
本文提出一种基于LPV模型的鲁棒迭代学习模型预测控制(Robust model predictive iterative learning control, RMPILC)算法, 实现间歇过程对变参考轨迹的跟踪.采用LPV模型描述非线性系统动态特性, 并通过状态增广建立二维误差模型.为保证变参考轨迹下的跟踪性能, 引入$H_\infty$约束条件.将变轨迹跟踪问题转化为LMI约束下的凸优化问题.通过针对数值例子以及CSTR系统的仿真验证了所提出算法的有效性.
1. 模型推导
1.1 非线性系统的LPV蕴含
假设非线性间歇系统由下式表示:
$ \left\{ \begin{array}{l} x(t+1)=f(x(t), u(t))\\ y(t)=g(x(t)) \\ \end{array} \right. $
(1) 其中, $x\in {\bf R}^{n_x}$是状态变量, $u\in{\bf R}^{n_u}$是控制输入, $y\in{\bf R}^{n_y}$为输出变量. $t\in(0, N]$, $N$为批次长度.
假设对任意$x(t)$、$u(t)$ $(t\in{(0, N]})$存在矩阵
$ \aleph(t)\in\Omega(\Im) $
满足
$ \left[\begin{array}{c} x(t+1)\\ y(t) \end{array} \right]=\aleph(t) {\left[\begin{array}{c} x(t)\\ u(t) \end{array} \right]} $
其中, $\Omega(\Im)=co\{A(\theta), B(\theta), C(\theta), {\theta}\in{\Im}\}$为多胞集合, 其中$\theta$为某一有界过程参数, 且存在$l$个非负系数$\theta_q~(q=1, 2, \cdots, l)$满足
$ A(\theta)=\sum\limits_{q=1}^l{{\theta_q}{A_q}}, B(\theta)=\sum\limits_{q=1}^l{{\theta_q}{B_q}}\\ C(\theta)=\sum\limits_{q=1}^l{{\theta_q}{C_q}}, \sum\limits_{q=1}^l{\theta_q}=1\\ $
那么, 非线性系统(1)的动态特性可由LPV系统描述:
$ \begin{cases} x(t+1)=A(\theta)x(t)+B(\theta)u(t)\\ y(t)=C(\theta)x(t)\\ \end{cases} $
(2) 即任何关于LPV系统(2)的性质适用于非线性系统(1).
间歇过程的LPV建模问题已经得到了广泛关注[25].其中, 选择合适的参数至关重要.对于简单非线性系统, 可以通过直接计算非线性项的上下界确定.若被控系统的非线性比较复杂, 可以基于系统平衡点, 采用数学变换的方法获得合适的表达式[26].
1.2 增广迭代误差模型
建立相邻迭代次序之间的动态关系, 得到
$ \left\{ \begin{array}{l} \Delta{x_{k}(t+1)}=A(\theta)\Delta{x_{k}(t)}+B(\theta)\Delta{u_{k}(t)}\\ \Delta{y_{k}(t)}=C(\theta)\Delta{x_{k}(t)}\\ \end{array} \right. $
(3) 其中
$ \Delta{x_{k}(t)}=x_{k}(t)-x_{k-1}(t)\\ \Delta{u_{k}(t)}=u_{k}(t)-u_{k-1}(t)\\ \Delta{y_{k}(t)}=y_{k}(t)-y_{k-1}(t) $
定义输出参考轨迹为$y{_k^r}(t)$, 那么跟踪误差可以定义为
$ e_{k}(t)=y{_k^r}(t)-y_{k}(t) $
(4) 将式(4)代入式(3), 得到沿迭代轴的增广迭代误差状态空间模型:
$ \begin{cases} \overline{x}_{k}(t+1)=\overline{A}(\theta) \overline{x}_{k}(t)+\overline{B}(\theta)\Delta{u_{k}(t)} +\Upsilon_{k}(t+1)\\ \overline{y}_{k}(t)=\overline{C}\overline{x}_{k}(t) \end{cases} $
(5) 其中
$ \begin{align*} &\overline{y}_{k}(t)=\Delta{e_{k}(t)},\overline{x}_{k}(t)= \left[\begin{array}{c} \Delta{x_{k}(t)}\\ \Delta{e_{k}(t)} \end{array} \right]\in{\bf R}^{n_x+n_y} \\& \Delta{e_{k}(t)}=e_{k}(t)-e_{k-1}(t),\Delta{y{_k^r}(t)}= y{_k^r}(t)-y{_{k-1}^r}(t) \\& \overline{A}(\theta)=\left[\begin{array}{cc} A(\theta)&0\\ -C(\theta){A}(\theta)&0 \end{array} \right]\\ &\overline{B}(\theta)=\left[\begin{array}{c} B(\theta)\\ -C(\theta){B}(\theta) \end{array} \right] \\& \overline{C}=\left[\begin{array}{cc} 0&I_{n_y\times{n_y}} \end{array} \right],\Upsilon_{k}(t+1)=\left[\begin{array}{c} 0\\ \Delta{y{_k^r}(t+1)} \end{array} \right] \end{align*} $
这里, $\Delta{y{_k^r}}$项代表了参考轨迹变化量.
1.3 二维增广误差模型
由第1.2节可知, 系统(5)的输出$\overline{y}_{k}(t)$为相邻迭代次序跟踪误差的变化量$\Delta{e_{k}(t)}$.则在第$k$次迭代中, 为了将跟踪误差$e_{k}(t)$控制到0, $\overline{y}_{k}(t)$的参考轨迹应为$\overline{y}{_k^r}(t)=-e_{k-1}(t)$.定义:
$ \tilde{e}_{k}(t)=\overline{y}{_k^r}(t)-\overline{y}_k(t) $
(6) 联立式(5)和式(6), 得到同时包含迭代域和时域动态特性的二维增广误差模型
$ \tilde{x}_{k}(t+1)=\tilde{A}(\theta)\tilde{x}_{k}(t)+\tilde{B}(\theta)\delta\Delta{u}_{k}(t)+R_{k}(t+1) $
(7) 其中
$\begin{align*} &\tilde{A}(\theta)=\begin{bmatrix} I_{n_y\times{n_y}}&-\tilde{C}\overline{A}(\theta)\\ 0&\overline{A}(\theta) \end{bmatrix}\nonumber\\&\tilde{B}(\theta)=\begin{bmatrix} -\tilde{C}\overline{B}(\theta)\\ \overline{B}(\theta) \end{bmatrix} \\& \tilde{x}_{k}(t)=\begin{bmatrix} -\tilde{e}_{k}(t)\\ \delta\overline{x}_{k}(t) \end{bmatrix},\delta\Delta{u}_{k}(t)=\Delta{u}_{k}(t)-\Delta{u}_{k}(t-1) \\& \delta\overline{x}_{k}(t)=\overline{x}_{k}(t)-\overline{x}_{k}(t-1) \\& R_{k}(t+1)=\begin{bmatrix} \delta\overline{y}{_k^r}(t+1)-\delta\Delta{y}{_k^r}(t+1)\\ \Upsilon_{k}(t+1)-\Upsilon_{k}(t) \end{bmatrix} \\& \delta\overline{y}{_k^r}(t+1)=\overline{y}{_k^r}(t+1)-\overline{y}{_k^r}(t) \\& \delta\Delta{y}{_k^r}(t+1)=\Delta{y}{_k^r}(t+1)-\Delta{y}{_k^r}(t) \end{align*} $
且满足
注1. $R_{k}(t+1)$包含了参考轨迹变化量, 为已知有界时变量, 与状态变量、控制输入变量均无关, 可以看作迭代域上的有界外部干扰.
注2. 模型(7)与典型的二维Rosser模型[27]不同, 它将在同一个状态方程中建立时域与迭代域上动态关系, 其中$\tilde{x}_{k}(t)$、$\delta\Delta{u}_{k}(t)$以及$R_{k}(t+1)$都是同时包含时域及迭代域信息的二维变量.
因此, 系统(1)的轨迹跟踪问题可以转化为系统(7)的零点跟踪问题.其控制任务包括:
1) 将(7)中的状态$\tilde{x}_{k}(t)$控制到0;
2) 限制参考轨迹变化量$R_{k}(t+1)$对控制性能的影响;
3) 防止控制输入波动过大.
2. RMPILC算法
2.1 问题描述
根据控制任务1)和3), 结合鲁棒$H_\infty$控制, 定义控制性能指标$z_k(t)\in{\bf R}^{n_x+2n_y+n_u}$
$ z_k(t)=C_{\infty}\tilde{x}_{k}(t)+D_{\infty}\delta\Delta{u}_{k}(t) $
(8) 其中
${\small\begin{align*} &C_\infty=\\ &\left[ \begin{array}{ccc} m_1 & & \\ & \ddots & \\ & & m_{n_x+2n_y} \\ \textbf{0}_{n_u\times1} & \cdots & \textbf{0}_{n_u\times1} \\ \end{array} \right]\in {\bf R}^{(n_x+2n_y+n_u)\times{n_x+2n_y}} \\ &D_{\infty}=\\ &\left[ \begin{array}{ccc} \textbf{0}_{(n_x+2n_y)\times1} & \cdots & \textbf{0}_{(n_x+2n_y)\times1} \\ n_1 & & \\ & \ddots & \\ & & n_{n_u} \\ \end{array} \right]\in{\bf R}^{(n_x+2n_y+n_u)\times{n_u}} \end{align*}} $
$m_i~(i=1, 2, \cdots, n_x+2n_y), n_j~(j=1, 2, \cdots, n_u)$为可调权重系数.显然, $C_\infty^{\rm T}{D_\infty}=0$.目标函数可表述为$z_k(t)$的二次型:
$ J{_k^\infty}(t)=\sum\limits_{i=0}^\infty{z_k^{\rm T}(t+i|t){z_k(t+i|t)}} $
(9) 为实现控制任务2), 引入$H_\infty$范数$\|T_{zR}\|{_\infty^2}$ :
$ \|T_{zR}\|{_\infty^2}=\frac{\sum\limits_{i=0}^{\infty}{\|z_k(t+i|t)\|^2}}{\sum\limits_{i=0}^ \infty{\|R_k(t+i+1)\|^2}} $
(10) 设置$H_\infty$性能指标
$ \|T_{zR}\|{_\infty^2}\leq\varepsilon $
(11) 其中, $\varepsilon>0$为给定的$H_\infty$性能上界.不等式(11)表示参考轨迹变化量$R_k(t+1)$对跟踪性能指标$z_k(t)$的影响被限制在由$\varepsilon$定义的范围内.
因此, 满足以上三项控制任务的优化问题可以描述为:
$ \underset{\delta\Delta{u}_{k}(t)}{\min\max} {J{_k^\infty}(t)} $
(12) 满足式(8)和式(12).
2.2 鲁棒稳定状态反馈控制律
定义状态反馈控制律
$ \delta\Delta{u}_{k}(t+i|t)=F_k(t)\tilde{x}_{k}(t+i|t), i\geq 0 $
(13) 其中, $\delta\Delta{u}_{k}(t+i|t)$、$\tilde{x}_{k}(t+i|t)$为当前时刻$t$对未来时刻$t+i$的预测值.
为表达简洁, 推导过程的书写省略$\theta$.考虑二次型函数$V(\tilde{x})=\tilde{x}^{\rm T}P\tilde{x}$, 可得
$ \begin{align} & V(\tilde{x}_{k}(t+i+1|t))-V(\tilde{x}_{k}(t+i|t))=\nonumber\\&\qquad \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \end{bmatrix}^{\rm T}\times\nonumber\\&\qquad \begin{bmatrix} (\tilde{A}+\tilde{B}F_{k}(t))^{\rm T}P(\tilde{A}+ \tilde{B}F_{k}(t))-P & \ast \\ P(\tilde{A}+\tilde{B}F_{k}(t)) & P \end{bmatrix}\times\nonumber\\&\qquad \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \end{bmatrix} \end{align} $
(14) 将式(14)从$i=0$累加至$i=\infty$, 可得
$\begin{align} &V(\tilde{x}_{k}(\infty|t))\!-\!V(\tilde{x}_{k}(t|t))\!=\! \sum_{i=0}^\infty \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix}^{\rm T}\!\!\times \nonumber\\&\qquad \begin{bmatrix} (\tilde{A}+\tilde{B}F_{k}(t))^{\rm T}P (\tilde{A}+\tilde{B}F_{k}(t))-P & \ast \\ P(\tilde{A}+\tilde{B}F_{k}(t)) & P \\ \end{bmatrix}\times\nonumber\\&\qquad \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix} \end{align} $
(15) 根据控制任务1), $\tilde{x}_k(\infty|t)$应为0, 即$V(\tilde{x}_k(\infty|t)=0$, 则
$ \begin{align} &-V(\tilde{x}_{k}(t|t))=\sum_{i=0}^\infty \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix}^{\rm T}\times\nonumber\\&\qquad \begin{bmatrix} (\tilde{A}+\tilde{B}F_{k}(t))^{\rm T}P(\tilde{A}+\tilde{B}F_{k}(t))-P & \ast \\ P(\tilde{A}+\tilde{B}F_{k}(t)) & P \\ \end{bmatrix}\times\nonumber\\&\qquad \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix} \end{align} $
(16) 由于$C_\infty^{\rm T}{D_\infty}=0$, 因此将式(8)代入式(9)可得
$ \begin{align} J{_k^\infty}(t)=\,&\varepsilon\sum_{i=0}^\infty{R_k^{\rm T}(t+i+1)R_k(t+i+1)}+ \nonumber\\&\sum_{i=0}^\infty \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix}^{\rm T} \begin{bmatrix} C_\infty^{\rm T}C_\infty & \ast \\ 0 & -\varepsilon{I} \\ \end{bmatrix}\times\nonumber\\& \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix} \end{align} $
(17) 联立式(16)和(17), 目标函数(9)有以下形式
$ J{_k^\infty}(t)= V(\tilde{x}_{k}(t|t))+\\ \varepsilon\sum\limits_{i=0}^\infty{R_k^{\rm T}(t+i+1)R_k(t+i+1)}+\\\sum\limits_{i=0}^\infty \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix}^{\rm T}\Phi \begin{bmatrix} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{bmatrix} $
(18) 其中
$ \begin{align*} \Phi=\begin{subarray}{l}\begin{bmatrix} (\tilde{A}+\tilde{B}F_{k}(t))^{\rm T}P(\tilde{A}+\tilde{B}F_{k}(t)) -P+&\ast\\C_\infty^{\rm T}C_\infty+F_k^{\rm T}(t)D_\infty^{\rm T}D_\infty{F_k(t)} & \\ P(\tilde{A}+\tilde{B}F_{k}(t))& P-\varepsilon{I}\\ \end{bmatrix}\end{subarray} \end{align*} $
引理1. 当且仅当$\Phi<0$, 能够满足$H_\infty$性能指标(11).
证明. 在式(14)两端同时加上$z_k^{\rm T}(t+i|t)z_k(t+i|t)-\varepsilon{R_k^{\rm T}(t+i+1)R_k(t+i+1)}$, 可得
$V(\tilde{x}(t+i+1|t))-V(\tilde{x}(t+i|t))=\\ \qquad -\|z_k(t+i|t\|^2+\varepsilon\|R_k(t+i+1)\|^2 +\\ \qquad\left[ \begin{array}{c} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{array} \right]^{\rm T}\Phi\left[ \begin{array}{c} \tilde{x}_{k}(t+i|t) \\ R_{k}(t+i+1) \\ \end{array} \right] $
(19) 若$\Phi<0$, 则有
$V(\tilde{x}_{k}(t+i+1|t))-V(\tilde{x}_{k}(t+i|t))\leq\\ \qquad -\|z_k(t+i|t)\|^2+\varepsilon\|R_k(t+i+1)\|^2 $
(20) 由于$V(\tilde{x}_{k}(\infty|t)=0$, $V(\tilde{x}_{k}(0|0)=0$, 将式(21)从$i=0$到$i=\infty$进行累加, 可得
$ \sum\limits_{i=0}^\infty{\|z_k(t+i|t)\|^2}\leq\varepsilon\sum\limits_{i=0}^\infty{\|R_k(t+i+1)\|^2} $
(21) 式(22)与$H_\infty$约束(12)等价.
在$\Phi<0$的条件下, 可以得到目标函数$J{_k^\infty}(t)$的上界
$ J{_k^\infty}(t)\leq {V(\tilde{x}_{k}(t|t))}+\\ \varepsilon\sum\limits_{i=0}^\infty{R_k^{\rm T}(t+i+1)R_k(t+i+1)} $
(22) 根据式(7)中$R_k(t+1)$的定义可知, $\sum_{i=0}^\infty{R_k^{\rm T}(t+i+1)R_k(t+i+1)}$为有界值.设
$\sum\limits_{i=0}^\infty{R_k^{\rm T}(t+i+1)R_k(t+i+1)}\leq{B_{R_k(t)}^2} \\ V(\tilde{x}_{k}(t|t))\leq\gamma $
(23) 联立式(22)和(23), 得
$ J{_k^\infty}(t)\leq\gamma+\varepsilon{B_{R_k(t)}^2} $
(24) 也就是说$J{_k^\infty}(t)$有上界$\gamma+\varepsilon{B_{R_k(t)}^2}$, 其中只有$\gamma$为变量.因此, 优化问题(12)可以改写为以最小化$\gamma$为优化目标, 以$F_{k}(t)$为优化变量的典型凸优化问题:
$ \underset{F_{k}(t)}{\min\max}{\gamma} $
(25a) 满足
$ V(\tilde{x}_{k}(t|t))\leq\gamma $
(25b) $ \Phi<0 $
(25c) 引理2. 若优化问题(25)在当前时刻可行, 当满足不等式
$ \tilde{x}_{k}^{\rm T}(t|t)P\tilde{x}_{k}(t|t)+\varepsilon{B_{R_k(t)}^2}\leq\gamma $
(26) 时, 由RMPILC算法控制的闭环系统是鲁棒稳定的.
证明. 联立式(14)及条件(26), 可得
$\tilde{x}_{k}^{\rm T}(t+i+1|t)P\tilde{x}_{k}(t+i+1|t)-\\ \qquad \tilde{x}_{k}^{\rm T}(t+i+1|t)P\tilde{x}_{k}(t+i+1|t)\leq\\ \qquad \varepsilon{R_k^{\rm T}(t+i+1)R_k(t+i+1)} $
(27) 将式(28)从$i=0$到$i=p-1~(p\in[1, \infty))$进行累加, 可得
$ \begin{array}{l} \tilde x_k^{\rm{T}}(t + p|t)P{{\tilde x}_k}(t + p|t) \le \tilde x_k^{\rm{T}}(t|t)P{{\tilde x}_k}(t|t) + \\ \varepsilon \sum\limits_{i = 0}^{p - 1} {R_k^{\rm{T}}(t + i + 1){R_k}(t + i + 1)} \end{array} $
(28) 联立式(23)、(26)和(28), 可以推出
$ \tilde{x}_{k}^{\rm T}(t+p|t)P\tilde{x}_{k}(t+p|t)\leq\gamma $
(29) 因此, 对于任意未来时刻$t+p$, 其状态$\tilde{x_k}(t+p|t)$属于不变集$\Omega_{\tilde{x_k}}$:
$ \Omega_{\tilde{x_k}}=\{x|x^{\rm T}\gamma^{-1}Px\leq1\} $
(30) RMPILC控制下的闭环系统是鲁棒稳定的.
注3. 若参考输出保持不变, 即$R_k(t+1)=0$, 那么RMPILC控制下的闭环系统是Lyapunov意义下稳定的.
证明. 根据式(27), 若$R_k(t+1)=0$, 能推出
$ \tilde{x}_{k}^{\rm T}(t+1|t)P\tilde{x}_{k}(t+1|t)\leq\tilde{x}_{k}^{\rm T}(t)P\tilde{x}_{k}(t) $
选择$V_k(t)=\tilde{x}_{k}^{\rm T}(t)P\tilde{x}_{k}(t)$作为Lyapunov函数, 可得到$V_k(t)$随时间衰减.因此, 闭环系统是Lyapunov稳定的.
2.3 LMI求解
为获得满足在鲁棒稳定条件(26)下优化问题(25)的最优解, 将其转化为线性矩阵不等式的形式.
引理3. 满足约束(25b)、(25c)及鲁棒稳定条件(26)的状态反馈矩阵$F_k(t)$可通过$F_k(t)=YQ^{-1}$计算得到, 其中$Q=\gamma P^{-1}$, $Y$为下述LMI约束下优化问题的解:
$ \underset{Y, Q}{\min}{\gamma} $
(31a) 对所有$q=1, 2, \cdots, l$, 满足
$ \left[ \begin{array}{ccccc} -Q & \ast & \ast & \ast & \ast \\ 0 & -\varepsilon\gamma I & \ast & \ast & \ast \\ \tilde{A}_qQ+\tilde{B}_qY & \gamma I & -Q & \ast & \ast \\ C_\infty Q & 0 & 0 & -\gamma I & \ast \\ D_\infty Y & 0 & 0 & 0 & -\gamma I \\ \end{array} \right] $
(31b) $ \left[ \begin{array}{ccc} 1 & \ast & \ast \\ \varepsilon{B_{R_k(t)}^2}& \gamma\varepsilon{B_{R_k(t)}^2} & \ast \\ \tilde{x}_{k}(t) & 0 & Q \\ \end{array} \right] $
(31c) 证明. 采用Schur补定理[28], 式(25c)等价于
$ \left[ \begin{array}{ccccc} -P & \ast & \ast & \ast & \ast \\ 0 & -\varepsilon I & \ast & \ast & \ast \\ \tilde{A}_qQ+\tilde{B}_qF_k(t) & I & -P^{-1} & \ast & \ast \\ C_\infty & 0 & 0 & -I & \ast \\ D_\infty & 0 & 0 & 0 & -I \\ \end{array} \right] $
(32) 分别左乘右乘diag$\{P^{-1}, I, I, I, I\}$, 得
$ \left[ \begin{array}{ccccc} -P^{-1} & \ast & \ast & \ast & \ast \\ 0 & -\varepsilon I & \ast & \ast & \ast \\ (\tilde{A}_qQ+\tilde{B}_qF_k(t))P^{-1} & I & -P^{-1} & \ast & \ast \\ C_\infty P^{-1} & 0 & 0 & -I & \ast \\ D_\infty P^{-1} & 0 & 0 & 0 & -I \\ \end{array} \right] $
(33) 将$P=\gamma Q^{-1}$, $F_k(t)=YQ^{-1}$代入式(33), 式(31b)可以被推出.
注意式(26)为式(25b)的充分条件, 也就是说只需要要满足式(26), 式(25b)也能被满足.将$P=\gamma Q^{-1}$代入式(27), 利用Schur补定理, 即可得到矩阵不等式(31c).
根据优化得到的$F_k(t)$, 通过下式计算控制输入$u_k(t)$:
$ u_k(t)= \delta\Delta{u_k(t)}+\Delta{u_k(t-1)}+u_{k-1}(t)=\\ F_k(t)\tilde{x}_{k}(t|t)+\Delta{u_k(t-1)}+u_{k-1}(t)=\\ YQ^{-1}\tilde{x}_{k}(t|t)+\Delta{u_k(t-1)}+u_{k-1}(t) $
(34) 其中, $\Delta{u_k(t-1)}$、$u_{k-1}(t)$为当前批次当前时刻的已知量. $\tilde{x}_{k}(t|t)$等于当前状态$\tilde{x}_{k}(t)$.
2.4 控制输入约束
间歇过程中需要考虑的控制输入约束包括$u_k(t)$、$\Delta{u_k(t)}$和$\delta{u_k(t)}$, 通常表述为
$ \left\{ \begin{array}{l} \|u_k(t)\|^2\leq u_h^2\\ \|\Delta u_k(t)\|^2\leq \Delta u_h^2\\ \|\delta u_k(t)\|^2\leq \delta u_h^2 \end{array} \right. $
(35) 推导$u_k(t)$, $\Delta{u_k(t)}$, $\delta{u_k(t)}$与$\delta\Delta{u_k(t)}$的关系
$ \left\{ \begin{array}{l} u_k(t)=\delta\Delta{u_k(t)}+\Delta{u_k(t-1)}+u_{k-1}(t)\\ \Delta u_k(t)=\delta\Delta{u_k(t)}+\Delta{u_k(t-1)}\\ \delta u_k(t)=\delta\Delta{u_k(t)}+\delta{u_{k-1}(t)} \end{array} \right. $
(36) 从式(36)可以看出, $u_k(t)$, $\Delta{u_k(t)}$, $\delta{u_k(t)}$都可以表示成$\delta\Delta{u_k(t)}$与其他已知量的和的形式.将其表述为以下通式
$ u_c=H\delta\Delta{u_k(t)}+u_m $
(37) 其中, $u_c$是被约束量, $u_m$是已知量, $H$是用于选择$\delta\Delta{u_k(t)}$中某一控制输入的向量.
式(35)中的约束条件可以统一表述为
$ \|u_c\|^2\leq\mu^2 $
(38) 其中, $\mu$代表约束上界.
结合式(37), 对不等式(38)进行放缩
$ \begin{align} & \|{{u}_{c}}{{\|}^{2}}=\|H\delta \Delta {{u}_{k}}(t)+{{u}_{m}}{{\|}^{2}}= \\ & \|H\delta \Delta {{u}_{k}}(t){{\|}^{2}}+2{{u}_{m}}H\delta \Delta {{u}_{k}}(t)+ \\ & u_{m}^{2}\le \|HY{{Q}^{-\frac{1}{2}}}{{Q}^{-\frac{1}{2}}}{{{\tilde{x}}}_{k}}(t|t){{\|}^{2}}+ \\ & 2{{u}_{m}}Y{{Q}^{-1}}{{{\tilde{x}}}_{k}}(t|t)+u_{m}^{2}\le \\ & \|HY{{Q}^{-\frac{1}{2}}}{{\|}^{2}}+2{{u}_{m}}Y{{Q}^{-1}}{{{\tilde{x}}}_{k}}(t|t)+u_{m}^{2}= \\ & \|HY{{Q}^{-\frac{1}{2}}}+{{u}_{m}}{{{\tilde{x}}}_{k}}{{(t|t)}^{\text{T}}}{{Q}^{-\frac{1}{2}}}{{\|}^{2}}+u_{m}^{2}- \\ & \|{{u}_{m}}{{Q}^{-\frac{1}{2}}}{{{\tilde{x}}}_{k}}(t|t){{\|}^{2}}= \\ & \|HY{{Q}^{-\frac{1}{2}}}+{{u}_{m}}{{{\tilde{x}}}_{k}}{{(t|t)}^{\text{T}}}{{Q}^{-\frac{1}{2}}}{{\|}^{2}}+ \\ & u_{m}^{2}(1-\beta )\le {{\mu }^{2}} \\ \end{align} $
(39) 其中, $0\leq\beta=\tilde{x}_{k}^{\rm T}(t|t)Q^{-1}\tilde{x}_{k}(t|t)$.由式(29)可知, $\tilde{x}_{k}^{\rm T}(t|t)Q^{-1}\tilde{x}_{k}(t|t)\leq1$.因此$\beta\in[0\;\;\;1]$.则式(39)可以写成以下线性矩阵不等式形式:
$ \begin{equation} \left[ \begin{array}{cc} \mu^2-u_m^2(1-\beta) & \ast \\ Y^{\rm T}H^{\rm T}+u_m\tilde{x}_{k}(t|t) & Q \\ \end{array} \right] \end{equation} $
(40) 因此控制输入约束下的鲁棒迭代预测控制优化问题可以描述为:
$ \underset{Y, Q}{\min}{\gamma} $
(41) 满足式(31b), 式(31c), 式(40).
3. 收敛性分析
MPILC控制系统的收敛性指的是当迭代次数趋近于无穷时, 跟踪误差收敛到零, 即对于任意$t\in[0, N]$, 当$k\rightarrow\infty$时, $e_k(t)\rightarrow0$.文献[29]在$x_{k-1}(N)=x_{k}(0)$的前提下, 证明了经典MPILC算法的收敛性.而近年研究中, 设计附加条件使$\|e_{k+1}(t)\|\leq a\|e_{k}(t)\|~(0<a<1)$成为保证收敛性更为常用的方法[9, 11].在本文提出的RMPILC算法中, 基于包含参考轨迹变化量的LPV模型(7), $\|e_{k+1}(t)\|\leq a\|e_{k}(t)\|$可以转化为有关增广状态$\tilde{x}_{k}(t)=[\tilde{e}_{k}(t)^{\rm T}\delta\overline{x}_k(t)^{\rm T}]^{\rm T}$的约束, 将其加入到实时优化中, 就可以保证原非线性系统在RMPILC控制下沿迭代轴的收敛性.
首先, 提出以下符合实际过程情况的假设:
1) 优化问题(31)在初始时刻可行;
2) 对于$t\in[0, N-1]$, 存在满足约束式(31b), 式(31c), 式(40)的控制序列$u_k(t)$令跟踪误差$e_k(t+1)$等于0;
3) 在$t$时刻, 已知$t+1$时刻的参考轨迹变化量.
定理1. 若在每个采样时刻的优化中, 对于$q=1, 2, \cdots, l$, $j=1, 2, \cdots, n_y$满足LMI约束
$ \left[ \begin{align} & {{a}^{2}}e_{k-1}^{j}{{(t+1)}^{2}}-(1+R_{k}^{\text{T}}(t+1)\times \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ * \\ & {{R}_{k}}(t+1)){{L}_{j}}L_{j}^{\text{T}} \\ & Q{{{\tilde{A}}}_{q}}L_{j}^{\text{T}}+{{Y}^{\text{T}}}\tilde{B}_{q}^{\text{T}}L_{j}^{\text{T}}\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \frac{Q}{1+R_{k}^{\text{T}}(t+1){{R}_{k}}(t+1)} \\ \end{align} \right] $
(42) 那么RMPILC控制下的跟踪误差沿迭代轴收敛到零.其中$e{_{k-1}^j}(t)$表示向量$e_{k-1}(t)$的第$j$个元素, $L_j$是用于选择向量$\tilde{e}_k(t)$第$j$个元素的给定向量.
证明. 在当前时刻$t$, 为保证下一时刻的跟踪误差沿迭代轴收敛, 即$\|e_{k+1}(t)\|\leq a\|e_{k}(t)\|$, 应满足以下不等式条件:
$ |e{_{k}^j}(t+1)|<a|e{_{k-1}^j}(t+1)| $
(43) 其中, $j=1, 2, \cdots, n_y$.
根据式(5)有
$ e{_{k}^j}(t)=e{_{k-1}^j}(t)+\Delta{e_k(t)} $
(44) 联立式(5)和(6)得
$ \Delta{e{_{k}^j}(t+1)}=-e{_{k}^j}(t+1)-\tilde{e}{_{k}^j}(t+1) $
(45) 联立式(43)、(44)和(45)可以得到
$ |\tilde{e}{_{k}^j}(t+1)|<a|e{_{k-1}^j}(t+1)| $
(46) 为不等式(43)的充分条件. (46)可以转化为系统(7)的状态约束:
$ \tilde{x}_k^{\rm T}(t+1)L_j^{\rm T}L_j\tilde{x}_k(t+1)<a^2e{_{k}^j}(t+1)^2 v $
(47) 为了得到式(47)的LMI表述, 进行以下推导[30]:
$\begin{align} &\tilde{x}_k^{\rm T}(t+1)L_j^{\rm T}L_j \tilde{x}_k(t+1) < a^2e{_{k-1}^j}(t+1)^2=\nonumber\\ &\qquad((\tilde{A}+\tilde{B}F_{k}(t)) \tilde{x}_k(t)+R_k(t+1))^{\rm T}L_j^{\rm T}L_j\times\nonumber\\ &\qquad((\tilde{A}+\tilde{B}F_{k}(t)) \tilde{x}_k(t)+R_k(t+1))=\nonumber\\ &\qquad\left[ \begin{array}{c} Q^{-\frac{1}{2}}\tilde{x}_k(t) \\ R_k(t+1)\\ \end{array} \right]^{\rm T} \left[ \begin{array}{c} Q^{\frac{1}{2}}(\tilde{A}+\tilde{B}F_{k}(t))^{\rm T} \\ I \\ \end{array} \right]L_j^{\rm T}L_j\times\nonumber\\&\qquad \left[ \begin{array}{c} Q^{\frac{1}{2}}(\tilde{A}+\tilde{B}F_{k}(t))^{\rm T} \\ I \\ \end{array} \right]^{\rm T}\left[ \begin{array}{c} Q^{-\frac{1}{2}}\tilde{x}_k(t) \\ R_k(t+1)\\ \end{array} \right] < \nonumber\\ &\qquad(1+R_k(t+1))^{\rm T}R_k(t+1)))\times\nonumber\\ &\qquad\lambda_{\max}\Bigg(L_j\left[ \begin{array}{c} Q(\tilde{A}+\tilde{B}F_{k}(t))^{\rm T} \\ I \\ \end{array} \right]^{\rm T}\left[ \begin{array}{cc} Q^{-1} & 0 \\ 0 & I \\ \end{array} \right]\nonumber\\ &\qquad\left[ \begin{array}{c} Q(\tilde{A}+\tilde{B}F_{k}(t))^{\rm T} \\ I \\ \end{array} \right]L_j^{\rm T}\Bigg) \end{align} $
(48) 联立式(47)和式(48)可得
$\begin{align} &a^2e{_{k-1}^j}(t+1)^2-\left(1+R_k^{\rm T}(t+1)R_k(t+1)\right) \times\nonumber\\& \qquad\Bigg(L_j\left[ \begin{array}{c} Q(\tilde{A}+\tilde{B}F_{k}(t))^{\rm T} \\ I \\ \end{array} \right]^{\rm T}\left[ \begin{array}{cc} Q^{-1} & 0 \\ 0 & I \\ \end{array} \right]\times\nonumber\\&\qquad\left[ \begin{array}{c} Q(\tilde{A}+\tilde{B}F_{k}(t))^{\rm T} \\ I \\ \end{array} \right]L_j^{\rm T}\Bigg)>0 \end{align} $
(49) 根据Schur补定理, 式(49)等价于式(42).因此, 如果在每次优化中满足LMI (42), 那么有$|e{_{k}^j}(t+1)|<a|e{_{k-1}^j}(t+1)|$, 即$\|e_{k+1}(t)\|\leq a\|e_{k}(t)\|$.所以, 对于任意$t\in[0, N]$, 当$k\rightarrow\infty$时, $e_k(t)\rightarrow0$.
4. 仿真研究
本节设计两组仿真实验以验证所提出的RMPILC在处理变轨迹跟踪问题方面的有效性.仿真1针对非线性数值系统, 侧重于对算法的理论分析和验证; 仿真2针对典型的间歇CSTR系统, 侧重于对RMPILC的实际应用效果检验.为进行对比, 同时设计经典MPILC算法的仿真实验, 其预测模型为:
$ {\pmb e}_k(t+m|t)={\pmb e}_k(t|t)-{G}^m(t)\Delta{\pmb u}{_k^m}(t) $
(50) 其中, $m$代表预测时域和控制时域大小, ${\pmb e}_k\in {\bf R}^{(n_y\times N)\times1}$,
$ \Delta{\pmb u}{_k^m}(t)=\left[ \begin{array}{ccc} \Delta u_k^{\rm T}(t) & \cdots & \Delta u_k^{\rm T}(t+m-1) \\ \end{array} \right]^{\rm T} $
$ {G}=\left[ \begin{array}{cccc} g_{1,0} & 0 & \cdots & 0 \\ g_{2,0} & g_{2,1} & \cdots & 0 \\ \vdots & \vdots & \ddots & 0 \\ \underbrace{g_{N,0}}_{G(0)} & \underbrace{\cdots}_{G(1)} & \cdots & \underbrace{g_{N,N-1}}_{G(N-1)} \\ \end{array} \right] $
$ {G}^m(t)=\left[ \begin{array}{ccc} G(t) & \cdots & G(t+m-1) \\ \end{array} \right] $
$g_{i, j}\in {\bf R}^{n_y(i)\times n_u(j)}$为$j$时刻施加的单位脉冲信号输入在$i$时刻的脉冲响应矩阵.通常矩阵${G}$可以通过沿参考轨迹进行线性化得到.经典MPILC算法的目标函数为:
$ J{_{\rm MPILC}^k}(t)=\frac{1}{2}\left\{\|{\pmb e}_k(t+m|t)\|_{Q_1}+\|\Delta{\pmb u}{_k^m}\|_{R_1}\right\} $
(51) 其中, $Q_1$和$R_1$为权重系数矩阵.
4.1 仿真1
考虑如下非线性数值系统
$ \begin{equation} \left\{ \begin{array}{l} x(t+1)=\left[ \begin{array}{cc} 0.5 & 0.125 \\ 0.125 & -0.65+0.15\sin{x_1} \\ \end{array} \right]\cdot\\ \qquad\qquad x(t)+\left[ \begin{array}{c} 0.01 \\ 0.07 \\ \end{array} \right] \\ y(t)=\left[ \begin{array}{cc} 1 & 0 \\ \end{array} \right] x(t) \\ \end{array} \right. \end{equation} $
(52) 控制输入约束为
$ \|u_k(t)\|^2\leq8^2, \|\Delta u_k(t)\|^2\leq0.6^2, \|\delta u_k(t)\|^2\leq1^2 $
(53) 由于$-1\leq\sin{x_1}\leq1$恒成立, 选择$\theta_1=\frac{\sin{x_1}-(-1)}{1-(-1)}$, $\theta_1=\frac{1-\sin{x_1}}{1-(-1)}$, 非线性系统(52)就可以被如式(2)的LPV模型包含, 其中$l=2$,
$ A_1=\left[ \begin{array}{cc} 0.5 & 0.125 \\ 0.125 & -0.5 \\ \end{array} \right],\quad A_2=\left[ \begin{array}{cc} 0.5 & 0.125 \\ 0.125 & -0.8 \\ \end{array} \right] $
根据式(7), 可以得到
$\begin{align*} &\tilde{A}_1=\left[ \begin{array}{cccc} 1 & 0.5 & 0.125 & 0 \\ 0 & 0.5 & 0.125 & 0 \\ 0 & -0.125 & -0.5 & 0 \\ 0 & -0.5 & -0.125 & 0 \\ \end{array} \right]\\&\tilde{A}_2=\left[ \begin{array}{cccc} 1 & 0.5 & 0.125 & 0 \\ 0 & 0.5 & 0.125 & 0 \\ 0 & -0.125 & -0.8 & 0 \\ 0 & -0.5 & -0.125 & 0 \\ \end{array} \right] \end{align*} $
仿真设置两种参考轨迹如图 1所示.第1批次到第4批次的目标参考轨迹为$y_{r_1}$; 从第5批次开始, 目标参考轨迹变为$y_{r_2}$.仿真时间为10分钟, 采样时间为0.025分钟.
批次长度为400.第1批次的控制输入为零向量. $H_\infty$性能上界$\varepsilon$设为10.初始状态为$x(0)=[0, 0]^{\rm T}$.收敛条件(42)作为每次优化的约束, 保证跟踪误差的收敛性, 其中$a=0.9$.权重矩阵取为
$\begin{align*} &C_\infty=\left[ \begin{array}{c} {\rm diag}\{1,0,0,0\} \\ \textbf{0}_{1\times4} \\ \end{array} \right]\\&D_\infty=\left[ \begin{array}{cccc} 0 & 0 & 0 & 0.0002 \\ \end{array} \right]^{\rm T} \end{align*} $
在每个采样时刻, 通过求解约束(42)下的LMI优化问题(41)得到变量$Y$, $Q$, 继而通过式$F_k(t)=YQ^{-1}$计算当前时刻的状态反馈矩阵$F_k(t)$.仿真分析中选择参考轨迹转折点第61个采样时刻来比较变轨迹前后的状态反馈矩阵变化, 其结果如表 1所示. 图 2和3为RMPILC的跟踪曲线及对应的控制输入曲线.
表 1 $F_k(t)$优化值Table 1 Optimized feedback control law批次($k$) $F_k(61)$ 2 [$-$46.7539 $-$24.0899 $-$5.0529 0.0000] 3 [$-$42.9654 $-$25.0475 $-$3.7597 0.0000] 4 [$-$57.4573 $-$29.2520 $-$5.4621 $-$0.0000] 5 [$-$16.9782 $-$7.8604 $-$1.2311 $-$0.0000] 6 [$-$37.0429 $-$26.9746 $-$3.0976 0.0000] 7 [$-$41.3123 $-$27.2625 $-$2.9534 $-$0.0000] 8 [$-$54.1913 $-$32.1226 $-$4.9777 0.0000] 在经典MPILC仿真中, 设置$Q_1=I_{400\times400}$, $R_1=0.00015I_{20\times20}$, $m=20$.其跟踪曲线如图 4所示.
比较图 2和图 4, 在参考轨迹保持不变的第1 $\sim$第4批次, RMPILC从第2批次就能够精确跟踪$y_{r_1}$, 而MPILC直到第4批次才能较好地跟踪${y_{r_1}}$.这是因为RMPILC采用了LPV模型来描述原系统的非线性特性, 避免出现模型失配问题, 从而获得了更快的收敛速度; 在参考轨迹变为$y_{r_2}$的第5 $\sim$第8批次, RMPILC能快速跟踪$y_{r_2}$, 而MPILC难以及时适应变参考轨迹, 需要经过几次迭代才能达到较好的跟踪效果.因而, RMPILC采用$H_\infty$控制有效抑制了变参考轨迹的影响.
图 5为RMPILC和经典MPILC控制下各批次跟踪误差均方差(Main square error, MSE)的变化情况. RMPILC控制下MSE沿迭代轴收敛到零, 且在参考轨迹变化的第5批次, 没有明显波动, 保持收敛趋势, 而MPILC控制下MSE出现较大波动.这证明了RMPILC在变参考轨迹下能够保证跟踪误差沿迭代轴收敛.
RMPILC的控制性能与参考轨迹变化程度以及$H_\infty$性能上界$\varepsilon$的大小有较大关系.由式(11)可知, $\varepsilon$越小越有利于增强抗干扰能力.而式(26)表明$\varepsilon$减小将导致可行域的缩小.当可行域缩小到不能包含当前状态时, 优化问题将无解.因此, 在选择$\varepsilon$时, 要根据实际需要权衡变轨迹适应能力和可行性问题.对于系统(52)能够保证可行性的最小$\varepsilon$值为5.8. 图 6为参考轨迹发生变化的第5批次中, RMPILC在$\varepsilon=5.8$、$\varepsilon=10$和$\varepsilon=15$时的跟踪情况, 表明随着$\varepsilon$增大, RMPILC跟踪性能下降. 图 7为$x(0)=[0.01, 0.05]^{\rm T}$, $\varepsilon$分别取值5.8、10和15时不变集$\Omega_{\tilde{x}_k}=\{x|x^{\rm T}Q^{-1}x\leq1\}$在原状态空间的象集.由于不变集的大小能够反映可行域的大小, 因此图 7表明初始可行域随$\varepsilon$减小而缩小.
图 6 RMPILC控制下第5批次当$\varepsilon=5.8$、$\varepsilon=10$和$\varepsilon=15$时的跟踪曲线Fig. 6 The tracking trajectories in the fifth batch when $\varepsilon=5.8, 10, 15$
图 7 RMPILC控制下$\varepsilon=5.8$、$\varepsilon=10$和$\varepsilon=15$时的不变集$\Omega_{\tilde{x}_k}$在原状态空间的象集Fig. 7 The image set of $\Omega_{\tilde{x}_k}$ when $\varepsilon=5.8, 10, 15$4.2 仿真2: CSTR系统
连续搅拌反应釜系统中进行恒定体积、放热、不可逆化学反应$A\rightarrow B$.其控制任务为重复跟踪给定的反应温度轨迹, 且生成物$B$的产品质量很大程度依赖于跟踪精度.因此, 采用MPILC方法控制CSTR系统能适应其生产过程的重复性, 并且提高产品质量.
CSTR系统具有以下非线性微分方程描述[31]:
$ \left\{ \begin{array}{l} \dot{C}_A=\frac{q}{V}(C_{Af}-C_A)-k_0\exp\left(\frac{-E} {RT}\right)C_A\\ \dot{T}=\frac{q}{V}(T_f-T)+\frac{-\Delta H}{\rho C_p}k_0\exp\left(\frac{-E}{RT}\right)C_A+ \\ \qquad\frac{UA}{V\rho C_p}(T_c-T) \end{array} \right. $
(54) 其中, 反应温度$T$ (K)为被控量, 冷却剂温度$T_c$ (K)为控制输入.其他参数的物理意义和取值见文献[31].
在间歇反应器控制中, 反应温度$T$的参考轨迹可能会由于调整进料浓度$C_A$、启动速度、批次时间长度等发生变化.为了验证RMPILC在适应频繁变化的参考轨迹的能力, 在仿真中设置三种不同的参考轨迹, 如图 8所示, 包括常规轨迹$y_{r_1}$、慢启动轨迹$y_{r_2}$以及快启动轨迹$y_{r_3}$. $y_{r_3}$中$T$上升较快, 有利于提高产量; $y_{r_2}$中$T$上升较慢, 后续反应更加平稳, 易于控制.在实际生产中可以根据不同的生产需求选择不同的参考轨迹.
图 8 CSTR反应温度$T$参考轨迹Fig. 8 The reference trajectories of CSTR reaction temperature $T$根据文献[32]介绍的替换法, 非线性系统(54)可以表述为如式(2)的LPV模型, 过程如下:
首先, 计算系统(54)的平衡点:
$ \{C{_A^{eq}}, T^{eq}, T{_c^{eq}}\}=\{0.5 {\rm mol} , 350 {\rm K}, 338 {\rm K}\} $
定义状态变量$[x_1, x_2]^{\rm T}=[C_A-C{_A^{eq}}, T-T^{eq}]$, 输入变量$u=T_c-T{_c^{eq}}$, 输出变量$y=x_2$.系统(54)能写成以下的状态方程形式:
$ \left\{ \begin{array}{l} \dot{x}=\left[ \begin{array}{c} \frac{q}{V}(C_{Af}-x_1-C{_A^{eq}})- \\k_0\exp\left(\frac{-E}{R(x_2+T^{eq})}\right) (x_1+C{_A^{eq}}) \\ \frac{q}{V}(T_f-x_2-T^{eq})+ \frac{-\Delta H}{\rho C_P}\times\\k_0\exp\left(\frac{-E} {R(x_2+T^{eq})}\right)\times\\ (x_1+C{_A^{eq}})+ \frac{UA}{V\rho C_p}(T_c-x_2-T^{eq}) \\ \end{array} \right] \\ y=x_2 \end{array} \right. $
(55) 在给定的输出参考轨迹中, 反应温度满足$\underline{T}\leq T\leq\overline{T}$, 也就是$\underline{T}-T^{eq}\leq x_2<\overline{T}-T^{eq}$.设
$ \underline{x}=\underline{T}-T^{eq}, \overline{x}=\overline{T}-T^{eq} $
定义
$ \begin{align} & {{\varphi }_{1}}({{x}_{2}})={{k}_{0}}\exp \left( \frac{-E}{R({{x}_{2}}+{{T}^{eq}})} \right) \\ & {{\varphi }_{2}}({{x}_{2}})={{k}_{0}}(\exp \left( \frac{-E}{R({{x}_{2}}+{{T}^{eq}})} \right)- \\ & \exp \left( \frac{-E}{R{{T}^{eq}}} \right))C_{A}^{eq}\frac{1}{{{x}_{2}}} \\ & {{\nu }_{1}}({{x}_{2}})={{\varphi }_{1}}({{x}_{2}})-\frac{1}{2}({{\varphi }_{1}}({{{\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle-}$}}{x}}}_{2}})+{{\varphi }_{1}}({{{\bar{x}}}_{2}})) \\ & {{\nu }_{2}}({{x}_{2}})={{\varphi }_{2}}({{x}_{2}})-\frac{1}{2}({{\varphi }_{1}}({{{\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle-}$}}{x}}}_{2}})+{{\varphi }_{1}}({{{\bar{x}}}_{2}})) \\ \end{align} $
选择LPV模型(2)中的参数$\theta$为
$ \theta_1=\frac{1}{2}\frac{\nu_1(x_2)-\nu_1(\underline{x}_2)} {\nu_1(\overline{x}_2)-\nu_1(\underline{x}_2)}, \theta_2= \frac{1}{2}\frac{\nu_1(\overline{x}_2)-\nu_1(x_2)} {\nu_1(\overline{x}_2)-\nu_1(\underline{x}_2)} \\ \theta_1=\frac{1}{2}\frac{\nu_2(x_2)-\nu_2(\underline{x}_2)} {\nu_2(\overline{x}_2)-\nu_2(\underline{x}_2)}, \theta_2= \frac{1}{2}\frac{\nu_2(\overline{x}_2)-\nu_2(x_2)} {\nu_2(\overline{x}_2)-\nu_2(\underline{x}_2)} $
那么系统(55)可以由如式(2)的LPV模型描述, 其多胞形的各顶点为
$\begin{align*} &A_1=\left[ \begin{array}{cc} 0.8227 & -0.00168 \\ 6.1233 & 0.9367 \\ \end{array} \right]\\& A_2=\left[ \begin{array}{cc} 0.9654 & -0.00182 \\ -0.6759 & 0.9433 \\ \end{array} \right] \\& A_3=\left[ \begin{array}{cc} 0.8895 & -0.00294 \\ 2.9447 & 0.9968 \\ \end{array} \right]\\&A_4=\left[ \begin{array}{cc} 0.8930 & -0.00062 \\ 2.7738 & 0.8864 \\ \end{array} \right]\\& B_1=\left[ \begin{array}{c} -0.000092 \\ 0.1014 \\ \end{array} \right],~ B_2=\left[ \begin{array}{c} -0.000097 \\ 0.1016 \\ \end{array} \right] \\&B_3=\left[ \begin{array}{c} -0.000157 \\ 0.1045 \\ \end{array} \right],~ B_4=\left[ \begin{array}{c} -0.000034 \\ 0.0986 \\ \end{array} \right] \\&C=\left[ \begin{array}{cc} 1 & 0 \\ \end{array} \right] \end{align*} $
仿真基于LPV模型设计控制律, 并将优化得到的控制输入施加都原非线性系统(54)中.
控制输入$T_c$的约束如下:
$ \|T_c\|^2\leq350^2, \|\Delta T_c\|^2\leq5^2, \|\delta T_c\|^2\leq2.5^2 $
(56) 仿真时间为12分钟(min), 采样时间为0.03分钟(min), 批次长度为400.设置初始参考轨迹为$y_{r_1}$, 在第6批次、第7批次分别变为$y_{r_2}$、$y_{r_3}$, 在第8批次变回$y_{r_1}$.批次1的初始控制输入为幅度为330 K的阶跃信号.初始状态为$[C_A, T]^{\rm T}=[0.7 {\rm mol/L}, 340 {\rm K}]^{\rm T}$. $H_\infty$性能上界选为$\varepsilon=20$.权重系数矩阵选取同仿真1.同样地, 收敛条件(42)在每次优化中作为约束$(a=0.9)$, 状态反馈矩阵$F_k(t)$由$F_k(t)=YQ^{-1}$计算得到, 各批次$F_k(200)$的优化值如表 2所示.
表 2 $F_k(t)$优化值Table 2 Optimized feedback control law批次$k$ $F_k(200)$ 2 [-7.8076 -12.6079 -7.9428 -0.0000] 3 [-8.4202 -12.9000 -8.2264 -0.0000] 4 [-7.8744 -12.6839 -7.9521 -0.0000] 5 [-8.9258 -13.1178 -8.4572 -0.0000] 6 [-9.7286 -13.2893 -9.0092 0.0000] 7 [-6.9490 -11.3713 -7.6883 0.0000] 8 [-7.5195 -12.4532 -8.0074 -0.0000] 9 [-7.7803 -12.6691 -7.9535 -0.0000] 在经典MPILC仿真中, 设置$m=10$, $Q_1=I_{400\times400}$, $R_1=I_{10\times10}$. 图 9和图 11为RMPILC和MPILC控制下的跟踪曲线.相应的RMPILC控制输入如图 10所示.可以看出RMPILC从第2批次开始就可以准确跟踪$y_{r_1}$, 且在批次6 $\sim$ 8能够及时跟踪变化轨迹.而MPILC直至批次5才能跟踪上$y_{r_1}$, 且在批次6 $\sim$ 8不能适应参考轨迹变化.因此与经典MPILC相比, 基于LPV模型的RMPILC快速跟踪变参考轨迹, 有利于提高CSTR的生产效率.
图 9 RMPILC控制下反应温度$T$参考轨迹跟踪曲线Fig. 9 The tracking trajectories for $T$ under RMPILC control
图 10 RMPILC控制下控制输入$T_c$轨迹Fig. 10 The trajectories of control input $T_c$ under RMPILC图 12为变轨迹下RMPILC和MPILC仿真中MSE随迭代次数的变化情况.显然, 变参考轨迹下RMPILC的跟踪误差沿迭代轴收敛, 而MPILC的跟踪误差发生较大波动, 会导致产品质量下降.
5. 结束语
本文针对具有重复特性的非线性间歇过程, 提出一种能跟踪变参考轨迹的鲁棒迭代学习模型预测控制.控制器设计基于包含被控系统非线性动态特性的LPV模型, 将LPV模型进行状态增广建立二维迭代误差模型.在鲁棒$H_\infty$预测控制框架下, 设置$H_\infty$性能上界, 并据此构建LMI约束下的目标函数优化问题.分析RMPILC系统的鲁棒稳定性和迭代收敛性, 将其充分条件作为约束加入每个采样时刻的优化中.仿真结果验证了RMPILC在快速跟踪变参考轨迹方面的优势, 表明采用基于LPV模型的RMPILC算法能显著减少迭代学习次数, 提高生产效率.
在实际生产中, 间歇过程的参考轨迹可能会发生剧烈变化.若要保证鲁棒稳定条件和迭代收敛条件, 可能会导致优化问题不可行, 或是可行域太小以至达不到期望的跟踪精度.因此, 未来的研究方向趋向于构建软约束RMPILC算法.在优化中引入松弛变量来放松约束, 允许控制输入在短时间内超出约束以达到扩大可行域的目的[33].基于软约束的RMPILC算法将能够适应剧烈变化的参考轨迹, 提高间歇过程控制的鲁棒性.
-
表 1 各种数据关联算法的比较
Table 1 Comparison of various data association algorithms
方法 优点 缺点 适用场景 NN/GNN 计算量小, 实时性较好, 工程实现简单 抗干扰能力弱, 在量测信息密度较大或环境因素过于复杂时, 性能较差 仅适用于信噪比较高, 目标密度较小的场合 PDA 计算量和存储量较小, 易于工程实现 在密集杂波或多目标环境中容易产生误跟或丢失等现象 适用于杂波环境下的单目标或跟踪门不重叠的多目标环境 JPDA 不需要任何关于目标和杂波的先验信息 计算量和存储量大, 实时性差, 工程实现困难, 量测数和目标数较大时存在组合爆炸现象 适用于密集多目标和多杂波、目标数目恒定已知的情况 MHT 在理想条件下是处理数据关联的最优算法 计算量大, 过于依赖目标和杂波的先验知识, 假设数量随量测数和目标数呈指数级增长 适用于密集多目标和多杂波、目标数未知且时变的情况 PMHT 批处理方法, 计算量随目标数的增长而呈线性增长 目标的后验概率函数可能会收敛到局部最大值, 算法对初始值比较敏感 适用于密集多目标和多杂波、目标数未知且时变的情况 
下载: 导出CSV
-
[1] 李双建, 徐从春. 论海洋的战略地位和现代海洋发展观. 经济研究导刊, 2012, (27): 256−259Li Shuang-Jian, Xu Cong-Chun. Discussion on the Ocean's strategic position and modern ocean development concept. Economic Research Guide, 2012, (27): 256−259 [2] 刘忠, 刘志坤, 罗亚松, 颜冰, 胡洪宁. 水下自组织网络及军事应用. 北京: 国防工业出版社, 2015. 8−10Liu Zhong, Liu Zhi-Kun, Luo Ya-Song, Yan Bing, Hu Hong-Ning. Underwater Self-Organizing Network and Military Applications. Beijing: National Defense Industry Press, 2015. 8−10 [3] 郭戈, 王兴凯, 徐慧朴. 基于声呐图像的水下目标检测、识别与跟踪研究综述. 控制与决策, 2018, 33(5): 906−922Guo Ge, Wang Xing-Kai, Xu Hui-Pu. Review on underwater target detection, recognition and tracking based on sonar image. Control and Decision, 2018, 33(5): 906−922 [4] Chen Z Y, Xu W. Joint passive detection and tracking of underwater acoustic target by beamforming-based Bernoulli filter with multiple arrays. Sensors, 2018, 18(11): 4022 doi: 10.3390/s18114022 [5] Sozer E M, Stojanovic M, Proakis J G. Underwater acoustic networks. IEEE Journal of Oceanic Engineering, 2000, 25(1): 72−83 doi: 10.1109/48.820738 [6] Ghafoor H, Noh Y. An overview of next-generation underwater target detection and tracking: An integrated underwater architecture. IEEE Access, 2019, 7: 98841−98853 doi: 10.1109/ACCESS.2019.2929932 [7] Isbitiren G, Akan O B. Three-dimensional underwater target tracking with acoustic sensor networks. IEEE Transactions on Vehicular Technology, 2011, 60(8): 3897−3906 doi: 10.1109/TVT.2011.2163538 [8] Luo J H, Han Y, Fan L Y. Underwater acoustic target tracking: A review. Sensors, 2018, 18(2): 112 doi: 10.3390/s18010112 [9] Luo H J, Wu K S, Ruby R, Liang Y Q, Guo Z W, Ni L M. Software-defined architectures and technologies for underwater wireless sensor networks: A survey. IEEE Communications Surveys & Tutorials, 2018, 20(4): 2855−2888 [10] Akyildiz I F, Pompili D, Melodia T. Underwater acoustic sensor networks: Research challenges. Ad Hoc Networks, 2005, 3(3): 257−279 doi: 10.1016/j.adhoc.2005.01.004 [11] Tuna G, Gungor V C. A survey on deployment techniques, localization algorithms, and research challenges for underwater acoustic sensor networks. International Journal of Communication Systems, 2017, 30(17): e3350 doi: 10.1002/dac.3350 [12] Akyildiz I F, Pompili D, Melodia T. Challenges for efficient communication in underwater acoustic sensor networks. ACM SIGBED Review, 2004, 1(2): 3−8 doi: 10.1145/1121776.1121779 [13] 李风华, 路艳国, 王海斌, 郭永刚, 张飞. 海底观测网的研究进展与发展趋势. 中国科学院院刊, 2019, 34(3): 321−330Li Feng-Hua, Lu Yan-Guo, Wang Hai-Bin, Guo Yong-Gang, Zhang Fei. Research progress and development trend of seafloor observation network. Bulletin of the Chinese Academy of Sciences, 2019, 34(3): 321−330 [14] O'Donnell J, Codiga D, Edwards C, Ullman D, Hebert D, Rice J, et al. Front Resolving Observational Network with Telemetry (FRONT). Connecticut Univ Groton Marine Sciences Inst, 2002 [15] 佟盛, 张信学. 美军水下网络中心战技术发展及设想. 舰船科学技术, 2009, 31(2): 155−159Tong Sheng, Zhang Xin-Xue. The technological development and imagine of US navy underwater network centric warfare. Ship Science and Technology, 2009, 31(2): 155−159 [16] Grund M, Freitag L, Preisig J, Ball K. The PLUSNet underwater communications system: Acoustic telemetry for undersea surveillance. In: Proceedings of OCEANS 2006. Boston, MA, USA: IEEE, 2006. 1−5 [17] Killeen T L. Ocean Observatories Initiative (OOI): Advanced research tools for the ocean sciences. Marine Technology Society Journal, 2010, 44(6): 15−17 doi: 10.4031/MTSJ.44.6.18 [18] 李建如, 许惠平. 加拿大“海王星” 海底观测网. 地球科学进展, 2011, 26(6): 656−661Li Jian-Ru, Xu Hui-Ping. NEPTUNE-Canada. Advances in Earth Science, 2011, 26(6): 656−661 [19] 申中寅. 日本海洋实时监测系统DONET简介. 国际地震动态, 2018, (7): 34−40Shen Zhong-Yin. A brief introduction to DONET in Japan. Recent Developments in World Seismology, 2018, (7): 34−40 [20] Person R, Miranda J M, Puillat I. ESONIM − European seafloor observatory network implementation model. Esonews, 2008, 1(3): 1−8 [21] 杨丽坤. 海洋强国战略的顶层设计研究. 大连海事大学学报(社会科学版), 2015, 14(2): 73−77Yang Li-Kun. The top-level design research of the strategy of maritime power. Journal of Dalian Maritime University (Social Science Edition), 2015, 14(2): 73−77 [22] 国务院关于印发国家重大科技基础设施建设中长期规划(2012−2030年). 中国科技投资, 2013, (7): 9State Council's plan for issuing national mid- and long-term plans for the construction of major science and technology infrastructure (2012−2030). China Venture Capital, 2013, (7): 9 [23] 房建孟. 全国海洋观测网规划(2014−2020年). 中国海洋年鉴, 2015. 41−46Fang Jian-Meng. National Marine observation network planning (2014−2020). China Marine Yearbook, 2015. 41−46 [24] 许惠平, 张艳伟, 徐昌伟, 李建如, 刘丁, 覃如府, 等. 东海海底观测小衢山试验站. 科学通报, 2011, 56(22): 1839−1845 doi: 10.1360/csb2011-56-22-1839Xu Hui-Ping, Zhang Yan-Wei, Xu Chang-Wei, Li Jian-Ru, Liu Ding, Qin Ru-Fu, et al. Coastal seafloor observatory at Xiaoqushan in the East China Sea. Chinese Science Bulletin, 2011, 56(22): 1839−1845 doi: 10.1360/csb2011-56-22-1839 [25] 彭晓彤, 周怀阳, 吴邦春, 吕枫, 吴正伟, 杨灿军, 等. 美国MARS海底观测网络中国节点试验. 地球科学进展, 2011, 26(9): 991−996Peng Xiao-Tong, Zhou Huai-Yang, Wu Bang-Chun, Lv Feng, Wu Zheng-Wei, Yang Can-Jun, et al. Test China node on monterey accelerated research system (MARS). Advances in Earth Sciences, 2011, 26(9): 991−996 [26] 《声学技术》编辑部. 中国科学院声学研究所首次实现我国深海潜标数据无线实时传输. 声学技术, 2017, (1): 87Editorial Department of “Acoustics Technology”. The institute of acoustics of the Chinese Academy of Sciences has realized the wireless real-time transmission of deep-sea submersible data for the first time. Technical Acoustics, 2017, (1): 87 [27] 央视新闻客户端. 我国将建国家海底科学观测网总投资超20亿元. 科技传播, 2017, 9(11): 8CCTV News Client. China will build a national seabed science observation network with a total investment of more than 2 billion Yuan. Public Communication of Science & Technology, 2017, 9(11): 8 [28] Teymorian A Y, Cheng W, Ma L R, Cheng X Z, Lu X C, Lu Z X. 3D underwater sensor network localization. IEEE Transactions on Mobile Computing, 2009, 8(12): 1610−1621 doi: 10.1109/TMC.2009.80 [29] Lin W L, Li D S, Tan Y, Chen J, Sun T. Architecture of underwater acoustic sensor networks: A survey. In: Proceedings of the 2008 1st International Conference on Intelligent Networks and Intelligent Systems. Wuhan, China: IEEE, 2008. 155−159 [30] 陈华炎. 基于水下无线传感器网络的高能效目标跟踪研究[博士学位论文], 浙江大学, 中国, 2018Chen Hua-Yan. Energy-Efficient Target Tracking based on UWSNs [Ph. D. dissertation], Zhejiang University, China, 2018 [31] Li L, Qiao G X, Zhao J, Yang L J, Imtiaz M N. Design of communication node in underwater acoustic network. In: Proceedings of the 2013 International Conference on Mechatronic Sciences, Electric Engineering and Computer (MEC). Shengyang, China: IEEE, 2013. 591−594 [32] Maqsood H, Javaid N, Yahya A, Ali B, Khan Z A, Qasim U. MobiL-AUV: AUV-aided localization scheme for underwater wireless sensor networks. In: Proceedings of the 10th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS-2016). Fukuoka, Japan: IEEE, 2016. 170−175 [33] Gong Z J, Li C, Jiang F. AUV-aided joint localization and time synchronization for underwater acoustic sensor networks. IEEE Signal Processing Letters, 2018, 25(4): 477−481 doi: 10.1109/LSP.2018.2799699 [34] Yang H M, Sikdar B. A mobility based architecture for underwater acoustic sensor networks. In: Proceedings of the 2008 IEEE Global Telecommunications Conference. New Orleans, LO, USA: IEEE, 2008. 1−5 [35] Lin Y, Tao H X, Tu Y, Liu T. A node self-localization algorithm with a mobile anchor node in underwater acoustic sensor networks. IEEE Access, 2019, 7: 43773−43780 doi: 10.1109/ACCESS.2019.2904725 [36] 黎作鹏, 蔡绍滨, 张菁, 高振国, 赵靖. 水声传感器网络节点定位技术综述. 小型微型计算机系统, 2012, 33(3): 442−447Li Zuo-Peng, Cai Shao-Bin, Zhang Jing, Gao Zhen-Guo, Zhao Jing. Survey on node localization technology of underwater acoustic sensor networks. Journal of Chinese Computer Systems, 2012, 33(3): 442−447 [37] 周宏仁, 敬忠良, 王培德. 机动目标跟踪. 北京: 国防工业出版社, 1991. 12−13Zhou Hong-Ren, Jing Zhong-Liang, Wang Pei-De. Tracking of Maneuvering Targets. Beijing: National Defense Industry Press, 1991. 12−13 [38] Singer R A. Estimating optimal tracking filter performance for manned maneuvering targets. IEEE Transactions on Aerospace & Electronic Systems, 1970, AES-6(4): 473−483 [39] Moose R L. An adaptive state estimation solution to the maneuvering target problem. IEEE Transactions on Automatic Control, 1975, 20(3): 359−362 doi: 10.1109/TAC.1975.1100961 [40] 周宏仁. 机动目标“当前”统计模型与自适应跟踪算法. 航空学报, 1983, 4(1): 73−86Zhou Hong-Ren. A “Current” statistical model and adaptive tracking algorithm for maneuvering targets. Acta Aeronautica et Astronautica Sinica, 1983, 4(1): 73−86 [41] Mehrotra K, Mahapatra P R. A jerk model for tracking highly maneuvering targets. IEEE Transactions on Aerospace and Electronic Systems, 1997, 33(4): 1094−1105 doi: 10.1109/7.624345 [42] Blom H A P, Bar-Shalom Y. The interacting multiple model algorithm for systems with Markovian switching coefficients. IEEE Transactions on Automatic Control, 1988, 33(8): 780−783 doi: 10.1109/9.1299 [43] 党建武. 水下多目标跟踪理论. 西安: 西北工业大学出版社, 2009. 38−43Dang Jian-Wu. Underwater Multi-target Tracking. Xi＇ an: Northwestern Polytechnical University Press, 2009. 38−43 [44] Xu J, Xu M Z, Zhou X Y. The bearing only target tracking of UUV based on cubature Kalman filter with noise estimator. In: Proceedings of the 36th Chinese Control Conference. Dalian, China, 2017. 5288−5293 [45] Wang X, Xu M X, Wang H B, Wu Y. Combination of interacting multiple models with the particle filter for three-dimensional target tracking in underwater wireless sensor networks. Mathematical Problems in Engineering, 2012, 2012: Article ID 829451 [46] Zhang J X, Liu M Q, Fan Z. Classify motion model via SVM to track underwater maneuvering target. In: Proceedings of the 2019 IEEE International Conference on Signal Processing, Communications and Computing. Dalian, China: IEEE, 2019. 1−6 [47] He C F, Wang Y Y, Chen C L, Guan X P. Target localization for a distributed SIMO sonar with an isogradient sound speed profile. IEEE Access, 2018, 6: 29770−29783 doi: 10.1109/ACCESS.2018.2843438 [48] Ramezani H, Jamali-Rad H, Leus G. Target localization and tracking for an isogradient sound speed profile. IEEE Transactions on Signal Processing, 2013, 61(6): 1434−1446 doi: 10.1109/TSP.2012.2235432 [49] 张铎. 基于水下无线传感器网络的自组织目标跟踪算法研究[博士学位论文], 浙江大学, 中国, 2019Zhang Duo. Self-Organizing Target Tracking based on UWSNs [Ph. D. dissertation], Zhejiang University, China, 2019 [50] Rao S K, Murthy K S L, Rajeswari K R. Data fusion for underwater target tracking. IET Radar, Sonar and Navigation, 2010, 4(4): 576−585 doi: 10.1049/iet-rsn.2008.0109 [51] Smith D, Singh S. Approaches to multisensor data fusion in target tracking: A survey. IEEE Transactions on Knowledge and Data Engineering, 2006, 18(12): 1696−1710 doi: 10.1109/TKDE.2006.183 [52] Luo R C, Yih C C, Su K L. Multisensor fusion and integration: Approaches, applications, and future research directions. IEEE Sensors Journal, 2002, 2(2): 107−119 doi: 10.1109/JSEN.2002.1000251 [53] Qiu J, Xing Z R, Zhu C S, Lu K F, He J L, Sun Y B, et al. Centralized fusion based on interacting multiple model and adaptive kalman filter for target tracking in underwater acoustic sensor networks. IEEE Access, 2019, 7: 25948−25958 doi: 10.1109/ACCESS.2019.2899012 [54] Hare J, Gupta S, Song J N. Distributed smart sensor scheduling for underwater target tracking. In: Proceedings of the 2014 Oceans. John's, NL, Canada: IEEE, 2014. 1−6 [55] Braca P, Goldhahn R, Ferri G, Lepage K D. Distributed information fusion in multistatic sensor networks for underwater surveillance. IEEE Sensors Journal, 2015, 16(11): 4003−4014 [56] Yu C H, Choi J W. Interacting multiple model filter-based distributed target tracking algorithm in underwater wireless sensor networks. International Journal of Control, Automation and Systems, 2014, 12(3): 618−627 doi: 10.1007/s12555-013-0238-y [57] Hare J Z, Gupta S, Song J N, Wettergren T A. Classification induced distributed sensor scheduling for energy-efficiency in underwater target tracking sensor networks. In: Proceedings of the 2017 Oceans. Anchorage, AK, USA: IEEE, 2017. 1−7 [58] 关欣, 何友, 衣晓. 带反馈信息的纯方位水下分布式融合算法. 系统仿真学报, 2003, 15(7): 947−949Guan Xin, He You, Yi Xiao. Bearings-only underwater distributed fusion algorithm with feedback information. Journal of System Simulation, 2003, 15(7): 947−949 [59] Han X Y, Liu M Q, Zhang S L, Zhang Q F. A multi-node cooperative bearing-only target passive tracking algorithm via UWSNs. IEEE Sensors Journal, 2019, 19(22): 10609−10623 doi: 10.1109/JSEN.2019.2931885 [60] Liu M Q, Zhao L J, Zhang S L. Delay-estimation-based asynchronous particle filtering for passive target tracking in underwater wireless sensor networks. In: Proceedings of the 36th Chinese Control Conference (CCC). Dalian, China: IEEE, 2017, 8929−8934 [61] 赵立佳. 基于水下传感网络异步量测的目标跟踪算法研究[硕士学位论文], 浙江大学, 中国, 2019Zhao Li-Jia. Target Tracking Algorithms based on Asynchronous Measurement of USNs [Master thesis], Zhejiang University, China, 2019 [62] Yan J, Zhang X N, Luo X Y, Sun Y L, Guan X P. AUV assisted asynchronous localization for underwater sensor networks. In: Proceeding of the 35th Chinese Control Conference (CCC). Chengdu, China: IEEE, 2016. 7291−7296 [63] Yan J, Zhang X N, Luo X Y, Wang Y Y, Chen C L, Guan X P. Asynchronous localization with mobility prediction for underwater acoustic sensor networks. IEEE Transactions on Vehicular Technology, 2018, 67(3): 2543−2556 doi: 10.1109/TVT.2017.2764265 [64] 朱光明. 异步水下传感器网络目标跟踪算法研究[博士学位论文], 浙江大学, 中国, 2015Zhu Guang-Ming. Research on Target Tracking Algorithms for Asynchronous Underwater Sensor Networks [Ph. D. dissertation], Zhejiang University, China, 2015 [65] Yan J, Zhao H Y, Pu B, Luo X Y, Chen C L, Guan X P. Energy-efficient target tracking with UASNs: A consensus-based Bayesian approach. IEEE Transactions on Automation Science and Engineering, 2020, 17(3): 1361−1375 [66] 薛锋, 刘忠, 曲毅. 基于Unscented粒子滤波的无序观测下水下无线传感器网络目标被动跟踪. 传感技术学报, 2007, 20(12): 2653−2658Xue Feng, Liu Zhong, Qu Yi. Target passive tracking based on unscented particle filter with out-of-sequence measurements in underwater wireless sensor networks. Chinese Journal of Sensors & Actuators, 2007, 20(12): 2653−2658 [67] Qiu W, Li L, Lei P Z, Wang Z L. Multiple targets tracking by using probability data association and cubature Kalman filter. In: Proceedings of the 10th International Conference on Wireless Communications and Signal Processing (WCSP). Hangzhou, China: IEEE, 2018. 1−5 [68] Qiu W, Wang W K, Zhuang Z Y, Lei P Z, Li L. Using ship radiated noise spectrum feature for data association in underwater target tracking. In: Proceedings of the 20th International Conference on Information Fusion (Fusion). Xi＇ an, China: IEEE, 2017. 1−5 [69] Yao Y, Zhao J H, Wu L N. Doppler data association scheme for multi-target tracking in an active sonar system. Sensors, 2019, 19(9): 2003 doi: 10.3390/s19092003 [70] Seget K, Schulz A, Heute U. Multi-hypothesis tracking and fusion techniques for multistatic active sonar systems. In: Proceedings of the 13th International Conference on Information Fusion. Edinburgh, UK: IEEE, 2010. 1−8 [71] Li X H, Willett P, Baum M, Li Y A. PMHT approach for underwater bearing-only multisensor−multitarget tracking in clutter. IEEE Journal of Oceanic Engineering, 2016, 41(4): 831−839 doi: 10.1109/JOE.2015.2506220 [72] Li X H, Li Y A, Yu J, Chen X, Dai M. PMHT approach for multi-target multi-sensor sonar tracking in clutter. Sensors, 2015, 15(11): 28177−28192 doi: 10.3390/s151128177 [73] Li X H, Zhao C X, Lu X F, Wei W. Underwater bearings-only multitarget tracking based on modified PMHT in dense-cluttered environment. IEEE Access, 2019, 7: 93678−93689 doi: 10.1109/ACCESS.2019.2927403 [74] 江同洋. 基于随机有限集的多目标贝叶斯滤波方法研究[博士学位论文], 浙江大学, 中国, 2015Jiang Tong-Yang. Multi-target Bayes Filtering based on Random Finite Sets [Ph. D. dissertation], Zhejiang University, China, 2015 [75] Mahler R P S. Multitarget Bayes filtering via first-order multitarget moments. IEEE Transactions on Aerospace and Electronic Systems, 2003, 39(4): 1152−1178 doi: 10.1109/TAES.2003.1261119 [76] Mahler R. PHD filters of higher order in target number. IEEE Transactions on Aerospace and Electronic Systems, 2007, 43(4): 1523−1543 doi: 10.1109/TAES.2007.4441756 [77] Melo J, Matos A. A PHD filter for tracking multiple AUVs. In: Proceedings of the 2014 Oceans. St. John's, NL, Canada: IEEE, 2014. 1−8 [78] Zhang S Z, Zhang X M, Yu Y. A forward-backward PHD smoother for tracking multiple AUVs. In: Proceedings of OCEANS. Shanghai, China: IEEE, 2016. 1−8 [79] Chen X, Li Y A, Li Y X, Yu J. PHD and CPHD algorithms based on a novel detection probability applied in an active sonar tracking system. Applied Sciences, 2018, 8(1): 36 [80] 王佰录. 基于随机集理论的分布式多传感器多目标跟踪技术研究[博士学位论文], 电子科技大学, 中国, 2018Wang Bai-Lu. Distributed Multi-Sensor Multi-Target Tracking in the Framework of Random Finite Sets [Ph. D. dissertation], University of Electronic Science and Technology, China, 2018 [81] Jing L Y, He C B, Huang J G, Ding Z. Energy management and power allocation for underwater acoustic sensor network. IEEE Sensors Journal, 2017, 17(19): 6451−6462 doi: 10.1109/JSEN.2017.2737229 [82] Yu C H, Lee K H, Choi J W, Seo Y B. Distributed single target tracking in underwater wireless sensor networks. In: Proceedings of the 2008 SICE Annual Conference. Tokyo, Japan: IEEE, 2008. 1351−1356 [83] Huang Y, Liang W, Yu H B, Xiao Y. Target tracking based on a distributed particle filter in underwater sensor networks. Wireless Communications and Mobile Computing, 2008, 8(8): 1023−1033 doi: 10.1002/wcm.660 [84] Hare J, Gupta S, Song J N. Distributed smart sensor scheduling for underwater target tracking. In: Proceedings of the 2014 Oceans. St. John's, NL, Canada: IEEE, 2014. 1−6 [85] Zhang Q, Zhang C J, Liu M Q, Zhang S L. Local node selection for target tracking based on underwater wireless sensor networks. International Journal of Systems Science, 2015, 46(16): 2918−2927 doi: 10.1080/00207721.2014.880199 [86] Sun Y L, Yuan Y Z, Li X L, Xu Q M, Guan X P. An adaptive sampling algorithm for target tracking in underwater wireless sensor networks. IEEE Access, 2018, 6: 68324−68336 doi: 10.1109/ACCESS.2018.2879536 [87] Zhang S L, Chen H Y, Liu M Q. Adaptive sensor scheduling for target tracking in underwater wireless sensor networks. In: Proceedings of the 2014 International Conference on Mechatronics and Control (ICMC). Jinzhou, China: IEEE, 2014. 55−60 [88] Chen H Y, Zhang S L, Liu M Q, Zhang Q F. An artificial measurements-based adaptive filter for energy-efficient target tracking via underwater wireless sensor networks. Sensors, 2017, 17(5): 971 doi: 10.3390/s17050971 [89] Zhang Q, Liu M Q, Zhang S L. Node topology effect on target tracking based on UWSNs using quantized measurements. IEEE Transactions on Cybernetics, 2015, 45(10): 2323−2335 doi: 10.1109/TCYB.2014.2371232 [90] Zhang S L, Chen H Y, Liu M Q, Zhang Q F. Optimal quantization scheme for data-efficient target tracking via UWSNs using quantized measurements. Sensors, 2017, 17(11): 2565 doi: 10.3390/s17112565 [91] Luo J H, Han Y, He X T. Optimal bit allocation for maneuvering target tracking in UWSNs with additive and multiplicative noise. Signal Processing, 2019, 164: 125−135 doi: 10.1016/j.sigpro.2019.06.005 [92] Zhang D, Liu M Q, Zhang S L, Fan Z, Zhang Q F. Mutual-information based weighted fusion for target tracking in underwater wireless sensor networks. Frontiers of Information Technology & Electronic Engineering, 2018, 19(4): 544−556 [93] Zhang D, Liu M Q, Zhang S L, Zhang Q F. Non-myopic energy allocation for target tracking in energy harvesting UWSNs. IEEE Sensors Journal, 2019, 19(10): 3772−3783 doi: 10.1109/JSEN.2018.2890078 [94] Han G J, Zhang C Y, Shu L, Rodrigues J J P C. Impacts of deployment strategies on localization performance in underwater acoustic sensor networks. IEEE Transactions on Industrial Electronics, 2015, 62(3): 1725−1733 doi: 10.1109/TIE.2014.2362731 [95] Zhang Q, Liu M Q, Zhang S L, Chen H Y. Node topology effect on target tracking based on underwater wireless sensor networks. In: Proceedings of the 17th International Conference on Information Fusion (FUSION). Salamanca, Spain: IEEE, 2014. 1−8 [96] Liu M Q, Zhang D, Zhang S L, Zhang Q F. Node depth adjustment based target tracking in UWSNs using improved harmony search. Sensors, 2017, 17(12): 2807 doi: 10.3390/s17122807 [97] Fariborzi H, Moghavvemi M. EAMTR: Energy aware multi-tree routing for wireless sensor networks. IET Communications, 2009, 3(5): 733−739 doi: 10.1049/iet-com.2008.0238 [98] Zhang R, Labrador M A. Energy-aware topology control in heterogeneous wireless multi-hop networks. In: Proceedings of the 2nd International Symposium on Wireless Pervasive Computing. San Juan, Puerto Rico: IEEE, 2007. 25−30 [99] Zeng D, Yan S F, Xu L J, Zhang Z. The low power DSP platform design of underwater acoustic communication system. In: Proceedings of the 2016 IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC). Hong Kong, China: IEEE, 2016. 1−4 [100] Jiang P, Liu J, Wu F, Wang J Z, Xue A K. Node deployment algorithm for underwater sensor networks based on connected dominating set. Sensors, 2016, 16(3): 388 doi: 10.3390/s16030388 [101] Bouabdallah F, Zidi C, Boutaba R. Joint routing and energy management in underwater acoustic sensor networks. IEEE Transactions on Network and Service Management, 2017, 14(2): 456−471 doi: 10.1109/TNSM.2017.2679482 [102] Chen K Y, Ma M D, Cheng E, Yuan F, Su W. A survey on MAC protocols for underwater wireless sensor networks. IEEE Communications Surveys & Tutorials, 2014, 16(3): 1433−1447 期刊类型引用(24)
1. 崔峥，王森，王娴，段浩，杨春曦，那靖. 基于颜色转移和自适应增益控制的混合水下图像增强. 自动化学报. 2025(02): 376-390 . 
本站查看2. 吴庆涛，朱军龙，葛泉波，张明川. 一种基于条件梯度的加速分布式在线学习算法. 自动化学报. 2024(02): 386-402 . 本站查看3. 徐齐胜，许可乐，窦勇，高彩丽，乔鹏，冯大为，朱博青. 基于被动声呐音频信号的水中目标识别综述. 自动化学报. 2024(04): 649-673 . 本站查看4. 张洪瑞，苏骏，李谦，李斌，寇小明. 基于主被动联合工作声呐探测信息的UKF目标运动要素解算方法. 水下无人系统学报. 2024(02): 354-361 . 百度学术5. 王浩亮，柴亚星，王丹，刘陆，王安青，彭周华. 基于事件触发机制的多自主水下航行器协同路径跟踪控制. 自动化学报. 2024(05): 1024-1034 . 本站查看6. 郑策，董超，郑兵，陈焱琨，贺惠忠. 面向方位历程交叉场景的多目标检测前跟踪方法. 声学学报. 2024(05): 990-1004 . 百度学术7. 田雨，申珅，赵罡，周景军，马宸宇浩，周大明. 基于信息融合的Singer模型机动目标跟踪算法研究. 舰船科学技术. 2024(24): 172-177 . 百度学术8. 王苏慧，王黎明，韩星程，叶泽甫，朱竹君. 基于最少Sigma点的水下动态目标定位算法改进研究. 舰船电子工程. 2023(01): 176-181 . 百度学术9. 尹堂文，孙艺洋，张晓霜. 虚拟物理人工智能多粒度水下态势感知研究. 指挥控制与仿真. 2023(03): 7-16 . 百度学术10. 毛琳琳，鄢社锋. 非理想水声信道下微弱信号分布式量化检测. 信号处理. 2023(07): 1203-1213 . 百度学术11. 田盛凯，张贞凯. 水下传感器网络目标跟踪时的非短视比特优化分配算法. 电讯技术. 2023(07): 1049-1056 . 百度学术12. 吕曜辉，胡明阳，唐胜雨，徐媛，李兴顺. 基于恒虚警检测的图像声呐目标跟踪. 数字海洋与水下攻防. 2023(04): 472-477 . 百度学术13. 刘志强，吴晓雄，沈廼桐，张旭. 基于移动WSN目标追踪算法. 计算机工程与设计. 2023(08): 2256-2262 . 百度学术14. 徐文，吴雨桑，张婷. 多阵列水下多目标跟踪的分布式算法研究. 信号处理. 2023(10): 1764-1774 . 百度学术15. 张延厚，王超，张奇，李强，笪良龙. 水声目标探测和识别融合技术发展综述. 信号处理. 2023(10): 1711-1727 . 百度学术16. 杜俊桥，夏伟杰，孙璟. 一种改进的CORDIC实时生成加权系数方法. 声学技术. 2023(05): 661-668 . 百度学术17. 郭瑛，张瑞，牛龙生，纪平. 水下传感器网络节点定位与轨迹预测技术研究进展. 青岛大学学报(自然科学版). 2022(01): 143-154 . 百度学术18. 喻俊志，孔诗涵，孟岩. 水下视觉环境感知方法与技术. 机器人. 2022(02): 224-235 . 百度学术19. 王百合，杨向锋，张群飞，雷开卓. 复杂水下战场环境中的多目标威胁评估及攻击优选方法. 火力与指挥控制. 2022(08): 177-182 . 百度学术20. 王苏慧，王黎明，韩星程，叶泽甫，朱竹君. 基于最大一致的水下目标协同定位. 机械与电子. 2022(12): 9-14 . 百度学术21. 马天力，张扬，刘盼，高嵩. 不确定重尾量测噪声干扰下的鲁棒目标跟踪算法. 空军工程大学学报. 2022(06): 64-70 . 百度学术22. 兰同宇，刘本奇，刘亮. 反UUV探测声呐系统发展综述. 水下无人系统学报. 2022(06): 704-713 . 百度学术23. 闫敬，关新平，罗小元，杨晛. 水下信息物理系统探测–通信–控制一体化:挑战与进展. 控制理论与应用. 2022(11): 1996-2008 . 百度学术24. 肖帅，张翰林，咸鹤群，陈飞. 一种面向物联网设备的口令认证密钥协商协议. 信息网络安全. 2021(10): 83-89 . 百度学术其他类型引用(33)
-
下载:
计量
- 文章访问数: 2961
- HTML全文浏览量: 1766
- PDF下载量: 966
- 被引次数: 57
