随着控制技术飞速发展,系统功能及其组件不断扩展,系统模型变得越来越复杂,工业控制需要建立更可靠的控制方法。然而,经典控制论的线性系统控制方法无法处理系统内存在的非线性、参数未知和系统模型部分未知等问题。因此,许多学者对非线性系统进行研究,并提出了自适应控制、模糊控制和神经网络控制等控制方法[1-7]。其中,自适应控制因其独特的参数更新方法受到学者们的重视。但是自适应控制需要获得被控对象的数学模型结构,因而难以处理建模困难且只能得到部分结构信息的非线性系统。近年来,随着神经网络技术的发展,神经网络的灵活性和对未知模型的逼近功能可以很好地弥补自适应控制的不足。因此,将神经网络与自适应控制相结合形成的自适应神经网络控制方法得到了广泛关注[8-10]。
对于大多数自适应神经网络控制方法,其控制效果很大程度上依靠控制器设计和各个参数的选取。为了使系统能够满足预期的性能要求,获得较小的稳态误差和较高的收敛速度,有关学者提出各类性能约束控制方法,其中包括障碍Lyapunov函数、漏斗控制和指定性能控制[11-18]。针对一类非严格反馈互联非线性系统,文献[11]提出了一种具有指定性能的自适应神经网络分散控制方法。针对多智能体系统,文献[17]通过设计指定性能函数,约束各个子系统的跟踪误差,从而得到期望的跟踪效果。基于四旋翼无人机动力学模型,文献[18]提出了一种自适应指定性能控制算法。另外,由于系统状态难以直接观测或观测成本高昂等问题,研究状态不可测系统具有非常重要的意义[19-23]。例如,文献[19]研究了一类具有状态不可测的非线性量化系统有限时间跟踪问题。文献[20]提出一种新的状态观测器,解决了一类纯反馈非线性系统的输出反馈控制问题。虽然学者们在非线性系统的自适应神经网络指定性能输出反馈控制方面已经取得了许多研究成果,但对于实际中执行器故障和全状态约束同时出现的情况,仍缺乏相关的分析和研究。
在元器件损耗、温湿度变化、外部扰动等因素的共同影响下,实际系统发生执行器故障的概率越来越大。执行器故障不仅会降低控制性能,还会使系统稳定性难以得到保障。因此,针对具有执行器故障的系统,其控制方法的研究具有重要的实际意义。文献[24]针对一类带有未知控制方向的多输入单输出执行器故障系统,提出了一种自适应输出反馈控制方法。文献[25]针对一类具有执行器故障和状态未知的不确定非线性互联大系统,将自适应Backstepping设计原理与Nussbaum增益函数特性相结合,提出了一种新的自适应神经网络输出反馈容错控制方法。另外,为了保证系统安全运行,系统状态需要满足一定的约束条件。文献[26]针对一类具有全状态约束的非线性系统,基于障碍李雅普诺夫函数,提出了一种自适应神经网络控制算法,使得系统状态满足约束要求,提高了系统的稳定性。基于对以上文献的分析,在自适应神经网络指定性能控制设计中考虑执行器故障和全状态约束具有重要意义。因此,本文针对一类具有执行器故障和全状态约束的纯反馈非线性系统,提出一种新的自适应神经网络输出反馈指定性能控制方法。
本文主要工作总结如下:(1) 相比于文献[17-18]的指定性能控制设计,本文可以通过设计性能函数中的参数来控制收敛时间和收敛速度,因此具有更高的灵活性。(2) 相比文献[17,26],本文采用非线性映射的方法,将约束系统转化为新的非约束系统,从而不需要对虚拟控制器进行有界的假设,降低系统保守性。(3) 本文考虑综合系统未知状态、指定性能、执行器故障和全状态约束的多约束条件下的一类非线性纯反馈系统,并解决其控制问题。此外,本文使用动态面技术避免了反步法中的“计算爆炸”问题,使控制设计的过程更加简单。
1 预备知识与问题描述 1.1 系统描述本文考虑如下的
$\left\{ \begin{array}{l} {\dot{x}}_{m}={f}_{m}\left({\overline{\boldsymbol{x}}}_{{m}{+1}}\right) +{x}_{m+1},1\leqslant m\leqslant n-1 \\ {\dot{x}}_{n}={f}_{n}\left({{\boldsymbol{x}}}\right) +{u}_{{\rm{f}}}\\ y={x}_{1} \end{array} \right. $ | (1) |
式中:
结合执行器可能存在的偏置故障和增益故障,本文所考虑系统的执行器故障模型描述为
$ {u}_{{\rm{f}}}\left(t\right) =b(1-s) u\left(t\right) +\varpi \left(t\right) $ |
式中:
在本文中,径向基函数神经网络(RBF NNs) 用于逼近系统中的非线性函数[17]
$ f\left({{\boldsymbol{Z}}}\right) ={\widehat{\boldsymbol{W}}}^{\mathrm{T}}\boldsymbol{S}({{\boldsymbol{Z}}}) $ |
式中:
$ f\left(\boldsymbol{Z}\right) ={\boldsymbol{W}}^{\mathrm{T}}\boldsymbol{S}\text{(}{{\boldsymbol{Z}}}\text{) }+\delta \left({{\boldsymbol{Z}}}\right) ,\forall {{\boldsymbol{Z}}}\in {\boldsymbol{\varOmega}} \subset {\bf{R}}^{q} $ |
式中:
$ {S}_{i}\left({{\boldsymbol{Z}}}\right) =\mathrm{e}\mathrm{x}\mathrm{p}\left[-\frac{{\left({{\boldsymbol{Z}}}-{{\boldsymbol{\iota}}}_{{i}}\right) }^{\mathrm{T}}\left({{\boldsymbol{Z}}}-{{\boldsymbol{\iota}}}_{{i}}\right) }{{\zeta }_{i}^{2}}\right] $ |
式中:
引理1[17] 令
$ {\|{{\boldsymbol{S}}}\left({\overline{\boldsymbol{x}}}_{{q}}\right) \|}^{2}\leqslant {\|{{\boldsymbol{S}}}\left({\overline{\boldsymbol{x}}}_{{p}}\right) \|}^{2} $ |
本文引入Nussbaum函数来处理未知增益问题,相关定义和引理如下。
定义1[27] 如果连续函数
$ \begin{array}{c}\underset{s\to \mathrm{\infty }}{\lim}\,{\rm{sup}}\dfrac{1}{s}{\displaystyle\int }_{0}^{s} N\left(\xi \right) {\rm{d}}\xi =+\infty \\ \underset{s\to \mathrm{\infty }}{\lim}\,{\rm{inf}}\dfrac{1}{s}{\displaystyle\int }_{0}^{s} N\left(\xi \right) {\rm{d}}\xi =-\infty \end{array} $ |
称为Nussbaum函数。
引理2[27] 设
$ 0 < V\left(t\right) \leqslant {c}_{0}+{{\rm{e}}}^{-{c}_{1}t}{\int }_{0}^{t} (g\left(x\left(\tau \right) \right) N\left(\xi \right) \\ + 1) \dot{\xi }{{\rm{e}}}^{{c}_{1}t}\mathrm{d}\tau $ |
式中:
本文控制目标如下:(1) 保证输出
为了便于状态观测器的设计,对系统(1) 进行如下变换:定义
$ \left\{ \begin{array}{l} {\dot{\chi }}_{m}=\dfrac{{f}_{m}\left({\overline{\chi}}_{{m}{+1}}\right) }{\beta }+{\chi }_{m+1} \\ {\dot{\chi }}_{n}=\dfrac{{f}_{n}\left(\boldsymbol{\chi}\right) }{\beta }+u+\dfrac{\varpi }{\beta }\\ \dot{y}=\beta {\chi }_{2}+{f}_{1}\left({\overline{\boldsymbol{x}}}_{\text{2}}\right) \end{array} \right. $ | (1) |
式中:
针对系统(1) ,考虑状态不可测问题,建立如下的状态观测器进行状态估计:
$ \left\{\begin{array}{l}{\dot{\hat{\chi }}}_{m}={\hat{\chi }}_{m+1}+{l}_{m}{e}_{1}\text{,}1\leqslant m\leqslant n-1\\ {\dot{\hat{\chi }}}_{n}=u+{l}_{n}{e}_{1}\end{array}\right. $ | (3) |
式中:
$ \dot{\boldsymbol{e}}=\boldsymbol{Ae}+\frac{\boldsymbol{F}{(}\boldsymbol{\chi}{) }}{\beta }+ {\boldsymbol{B}} \frac{\varpi }{\beta } $ | (4) |
式中:
由于
$ {f}_{m}={\boldsymbol{W}}_{{0}{m}}^{\text{T}}{\boldsymbol{S}}_{{0}}\left({{\boldsymbol{Z}}}\right) +{\delta }_{0m}\left({{\boldsymbol{Z}}}\right) ,{\delta }_{0m}\left({{\boldsymbol{Z}}}\right) \leqslant {\overline{\varepsilon }}_{0m}$ | (5) |
式中:
$ \boldsymbol{F}(\boldsymbol{\chi}) ={\boldsymbol{W}}_{{0}}^{\text{T}}{\boldsymbol{S}}_{{0}}({{\boldsymbol{Z}}}) +{\boldsymbol{\delta}}_{{0}}({{\boldsymbol{Z}}}) ,||{\boldsymbol{\delta}}_{{0}}({{\boldsymbol{Z}}}) ||\leqslant {\overline{\varepsilon }}_{0} $ |
式中:
$ \dot{\boldsymbol{e}}=\boldsymbol{Ae}+\frac{1}{\beta }\left({\boldsymbol{W}}_{{0}}^{\text{T}}{\boldsymbol{S}}_{{0}}+{\boldsymbol{\delta }}_{0}\right) +\boldsymbol{B}\frac{\varpi }{\beta } $ | (6) |
选取如下李雅普诺夫函数
$ {V}_{{\boldsymbol{e}}}={\boldsymbol{e}}^{\text{T}}\boldsymbol{Pe} $ | (7) |
结合式(6) ,可得
${\dot{V}}_{{\boldsymbol{e}}}=-\nu {\boldsymbol{e}}^{\text{T}}\boldsymbol{e}+\frac{2}{\beta }{\boldsymbol{e}}^{\mathrm{T}}\boldsymbol{P}\left({\boldsymbol{W}}_{{0}}^{\text{T}}{\boldsymbol{S}}_{{0}}+{\boldsymbol{\delta}}_{{0}}\right) +2{\boldsymbol{e}}^{\mathrm{T}}\boldsymbol{P}\boldsymbol{B}\frac{\varpi }{\beta } $ | (8) |
定义自适应参数为
$ {\theta }^{\mathrm{*}}=\mathrm{m}\mathrm{a}\mathrm{x}\left\{n{||{\boldsymbol{W}}_{{0}}||}^{2},{||{\boldsymbol{W}}_{{m}}||}^{2},1\leqslant m\leqslant n-1\right\} $ |
式中:
$ \frac{2}{\beta }{\boldsymbol{e}}^{\mathrm{T}}\boldsymbol{P}({\boldsymbol{W}}_{{0}}^{\text{T}}{\boldsymbol{S}}_{{0}}+{\boldsymbol{\delta}}_{{0}}) \leqslant 2\eta {\boldsymbol{e}}^{\text{T}}\boldsymbol{e}+{\eta }^{-1}\parallel \boldsymbol{P}{\parallel }^{2}{\theta }^{\mathrm{*}} +{\eta }^{-1}{\overline{\varepsilon }}_{0}^{2} $ | (9) |
$ \frac{2}{\beta }{\boldsymbol{e}}^{\text{T}}\boldsymbol{P}\boldsymbol{B}\varpi \leqslant \frac{1}{\varrho }{\boldsymbol{e}}^{\text{T}}\boldsymbol{e}+\varrho \parallel \boldsymbol{P}{\parallel }^{2}{\overline{\varpi }}^{2}$ | (10) |
式中:
将式(9) 和式(10) 代入到式(8) 中,可得
$ {\dot{V}}_{{\boldsymbol{e}}}\leqslant -\left(\nu -2\eta -\frac{1}{\varrho }\right) {\boldsymbol{e}}^{\text{T}}\boldsymbol{e}+{\eta }^{-1}\parallel \boldsymbol{P}{\parallel }^{2}{\theta }^{*}+ \varrho \parallel \boldsymbol{P}{\parallel }^{2}{\overline{\varpi }}^{2}+{\eta }^{-1}{\overline{\varepsilon }}_{0}^{2} $ | (11) |
本节将采用非线性映射进行全状态约束,并通过指定性能设计使跟踪误差满足期望的性能要求,进而设计具有全状态约束和指定性能的非线性系统的控制器和自适应律。
参考文献[28],针对系统(1) 的全状态约束问题,对状态观测器(3) 进行如下坐标转换:
$ \left\{ \begin{array}{l} {s}_{2}=\ln \dfrac{{\underline{K}}_{2}+{\hat{\chi }}_{2}}{{\overline{K}}_{2}-{\hat{\chi }}_{2}}\\ \qquad\quad \vdots \\ {s}_{n}={\ln}\dfrac{{\underline{K}}_{n}+{\hat{\chi }}_{n}}{{\overline{K}}_{n}-{\hat{\chi }}_{n}} \end{array} \right. $ | (12) |
式中:
注 1 该方法通过坐标转换将约束系统转化为无约束系统,再在新系统上进行控制器设计。相比于文献[17,26]中用障碍李雅普诺夫函数的全状态约束,不仅不需要对虚拟控制器进行有界的假设,降低保守性,并且其过程相对简单,降低了对系统进行状态约束控制设计的难度。
指定性能描述如下:
$ -\mu \left(t\right) < {\stackrel{~}{x}}_{1}\left(t\right) < \mu \left(t\right) ,\forall t > 0 $ |
式中:性能函数
$ \mu \left(t\right) =\left\{ {\begin{array}{*{20}{l}} \mathrm{l}\mathrm{n}({C}_{0}{\left({t}_{f}-t\right) }^{\overline{n}}+1) +{\mu }_{\mathrm{\infty }},& 0 < t < {t}_{{\rm{f}}}\\ {\mu }_{\mathrm{\infty }},& t\geqslant {t}_{{\rm{f}}}\end{array}} \right. $ |
式中:
注 2 相比于文献[15-16],本文所采用的性能函数可以通过设置参数
参考文献[29],本文选取如下误差转换函数:
$ {\xi }_{1}\left(t\right) =\frac{{\tilde{x}}_{1}}{\sqrt{{\mu }^{2}-{{\tilde{x}}_{1}}^{2}}} $ | (13) |
注 3 相比于多数指定性能设计使得
对
$ {\dot{\xi }}_{1}\left(t\right) ={{\varGamma }}_{1}\left({\dot{x}}_{1}-\frac{{\tilde{x}}_{1}\dot{\mu }}{\mu }\right) ={{\varGamma }}_{1}\left(\beta \left({s}_{2}+{e}_{2}\right) +{F}_{1}-{\dot{y}}_{d}-\frac{{\tilde{x}}_{1}\dot{\mu }}{\mu }\right) $ | (14) |
式中:
定义如下坐标变换:
$\begin{split} & {z}_{m}={s}_{m}-{\omega }_{m}\\& {\lambda }_{m}={\omega }_{m}-{\alpha }_{m-1},2\leqslant m\leqslant n \end{split} $ | (15) |
式中:
结合式(3)、式(12)、式(14) 和式(15) ,可得
$ \begin{split} & {\dot{\xi }}_{1}={{\varGamma }}_{1}\left(\beta \left({s}_{2}+{e}_{2}\right) +{F}_{1}-{\dot{y}}_{d}-\frac{{\tilde {x}}_{1}\dot{\mu }}{\mu }\right) \\& {\dot{z}}_{m}={Q}_{m}\left({G}_{m}+{s}_{m+1}+{l}_{m}{e}_{1}\right) -{\dot{\omega }}_{m} \\&{\dot{z}}_{n}={Q}_{n}\left(u+{l}_{n}{e}_{1}\right) -{\dot{\omega }}_{n} \end{split}$ | (16) |
式中:
自适应容错控制算法的设计步骤如下:
Step1:选取第一步的Lyapunov函数
$ {V}_{1}=\frac{1}{2}{\xi }_{1}^{2}+\frac{1}{2{r}_{0}}{\tilde \theta ^2} $ | (17) |
式中:
$ \begin{split} {\dot{V}}_{1}=&{\xi }_{1}{\dot{\xi }}_{1}-\frac{1}{{r}_{0}}\tilde \theta \dot{\hat{\theta }} ={\xi }_{1}{{\varGamma }}_{1}\Bigg(\beta \left({z}_{2}+{\lambda }_{2}+{\alpha }_{1}+{e}_{2}\right) + \\& {F}_{1}-{\dot{y}}_{d}-\frac{{\tilde{x}}_{1}\dot{\mu }}{\mu }\Bigg) -\frac{1}{{r}_{0}}\tilde \theta \dot{\hat{\theta }} \end{split}$ | (18) |
利用RBF NNs逼近
$ \begin{split} {\xi }_{1}{{\varGamma }}_{1}{F}_{1}=&{\xi }_{1}{{\varGamma }}_{1}\left({\boldsymbol{W}}_{{1}}^{\text{T}}{\boldsymbol{S}}_{{1}}\left({{\boldsymbol Z}}_{{1}}\right) +\delta \left({{\boldsymbol Z}}_{\text{1}}\right) \right) \leqslant \\& \frac{{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{\theta }^{*}{\boldsymbol{S}}_{{1}}^{\text{T}}\left({{\boldsymbol Z}}_{{1}}\right) {\boldsymbol{S}}_{{1}}\left({{\boldsymbol Z}}_{{1}}\right) }{2{a}_{1}^{2}}+\frac{{a}_{1}^{2}}{2}+\frac{{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}}{2}+\frac{{\overline{\varepsilon }}_{1}^{2}}{2}\leqslant \\& \frac{{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{\theta }^{*}{\boldsymbol{S}}_{{1}}^{\text{T}}\left({x}_{1}\right) {\boldsymbol{S}}_{{1}}\left({x}_{1}\right) }{2{a}_{1}^{2}}+\frac{{a}_{1}^{2}}{2}+\frac{{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}}{2}+\frac{{\overline{\varepsilon }}_{1}^{2}}{2} \end{split} $ | (19) |
${\xi }_{1}{{\varGamma }}_{1}\beta \left({z}_{2}+{\lambda }_{2}\right) \leqslant {\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{\overline{\beta }}^{2}+\frac{{\lambda }_{2}^{2}}{2}+\frac{{z}_{2}^{2}}{2}\;\;\; $ | (20) |
${\xi _1}{\varGamma _1}\beta {e}_{2}\leqslant \frac{{\xi }_{1}^{2}\varGamma _1^{2}{\overline \beta }^{2}}{2}+\frac{1}{2}\parallel {{\boldsymbol{e}}}{\parallel }^{2}\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; $ | (21) |
式中:
将式(19)~(21) 代入式(18) ,可得
$ \begin{split} & {\dot{V}}_{1}\leqslant \frac{{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{\theta }^{*}{||{\boldsymbol{S}}_{{1}}\left({x}_{1}\right) ||}^{2}}{2{a}_{1}^{2}}+\frac{1}{2}{a}_{1}^{2}+\frac{{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}}{2}+ \frac{1}{2}{\overline{\varepsilon }}_{1}^{2}+\frac{3{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{\overline{\beta }}^{2}}{2}+\\&\qquad \frac{{\lambda }_{2}^{2}}{2}+ \frac{{z}_{2}^{2}}{2}+\frac{1}{2}\parallel \boldsymbol{e}{\parallel }^{2}+{\xi }_{1}{{\varGamma }}_{1}\beta {\alpha }_{1}- {\xi }_{1}{{\varGamma }}_{1}\left({\dot{y}}_{{\rm{d}}}+\frac{{\tilde{x}}_{1}\dot{\mu }}{\mu }\right) -\frac{1}{{r}_{0}}\tilde{\theta }\dot{\hat{\theta }} \end{split}$ | (22) |
本文所选Nussbaum函数为
$ \begin{split} & {\alpha }_{1}=N\left(\psi \right) \Bigg({c}_{1}{\xi }_{1}+\frac{{\xi }_{1}{{\varGamma }}_{1}^{2}}{2}+\frac{{\xi }_{1}{{\varGamma }}_{1}^{2}\hat{\theta }{||{\boldsymbol{S}}_{{1}}\left({x}_{1}\right) ||}^{2}}{2{a}_{1}^{2}}+\Bigg.\\&\qquad \Bigg.\frac{3{\xi }_{1}{{\varGamma }}_{1}^{2}{\overline{\beta^{2} }}}{2}-{{\varGamma }}_{1}\left({\dot{y}}_{{\rm{d}}}+\frac{{\stackrel{~}{x}}_{1}\dot{\mu }}{\mu }\right) \Bigg) \\& \dot{\psi }={\xi }_{1}\Bigg({c}_{1}{\xi }_{1}+\frac{{\xi }_{1}{{\varGamma }}_{1}^{2}}{2}+\frac{{\xi }_{1}{{\varGamma }}_{1}^{2}\hat{\theta }{||{\boldsymbol{S}}_{{1}}\left({x}_{1}\right) ||}^{2}}{2{a}_{1}^{2}}+\Bigg.\\&\qquad \Bigg.\frac{3{\xi }_{1}{{\varGamma }}_{1}^{2}{\overline{\beta^{2} }}}{2}-{{\varGamma }}_{1}\left({\dot{y}}_{{\rm{d}}}+\frac{{\stackrel{~}{x}}_{1}\dot{\mu }}{\mu }\right) \Bigg) \end{split} $ | (23) |
将式(23)代入式(22) ,可得
$ \begin{split}& {\dot{V}}_{1}\leqslant -{c}_{1}{\xi }_{1}^{2}+\left({{\varGamma }}_{1}\beta N\left(\psi \right) +1\right) \dot{\psi }+ \frac{1}{{r}_{0}}\tilde{\theta }\left(\frac{{r}_{0}{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{∥{\boldsymbol{S}}_{\text{1}}\left({x}_{1}\right) ∥}^{2}}{2{a}_{1}^{2}}-\dot{\hat{\theta }}\right) +\\&\qquad \frac{{\overline{\varepsilon }}_{1}^{2}}{2}+\frac{{\lambda }_{2}^{2}}{2}+\frac{{z}_{2}^{2}}{2}+\frac{{a}_{1}^{2}}{2}+\frac{\parallel \boldsymbol{e}{\parallel }^{2}}{2} \end{split}$ | (24) |
Step m
$ {V}_{m}={V}_{m-1}+\dfrac{{z}_{m}^{2}}{2} $ | (25) |
结合式(16),对
$ \begin{split} {\dot{V}}_{m}=&{\dot{V}}_{m-1}+{z}_{m}{Q}_{m}\left({G}_{m}+{z}_{m+1}+{\lambda }_{m+1}+\right.\\& \left.{\alpha }_{m}+{l}_{m}{e}_{1}\right) -{z}_{m}{\dot{\omega }}_{m} \leqslant {\dot{V}}_{m-1}+{z}_{m}{Q}_{m}{G}_{m}+\\& \frac{3}{2}{z}_{m}^{2}{Q}_{m}^{2}+\frac{{z}_{m+1}^{2}}{2}+ \frac{{\lambda }_{m+1}^{2}}{2}+\frac{{l}_{m}^{2}}{2}\parallel \boldsymbol{e}{\parallel }^{2}-\\& {z}_{m}{\dot{\omega }}_{m}+{z}_{m}{Q}_{m}{\alpha }_{m} \end{split} $ | (26) |
由Young’s不等式和引理1,可得
$ {z}_{m}{Q}_{m}{G}_{m}\leqslant \frac{{z}_{m}^{2}{Q}_{m}^{2}{\theta }^{*}{||{\boldsymbol{S}}_{{m}}\left({\overline{\boldsymbol{s}}}_{{m}}\right) ||}^{2}}{2{a}_{m}^{2}}+\frac{1}{2}{a}_{m}^{2}+ \frac{{z}_{m}^{2}{Q}_{m}^{2}}{2}+\frac{1}{2}{\overline{\varepsilon }}_{m}^{2} $ | (27) |
式中:
$ {\alpha }_{m}=\frac{1}{{Q}_{m}}\left(-{c}_{m}{z}_{m}-\frac{{z}_{m}{Q}_{m}^{2}\hat{\theta }{||{\boldsymbol{S}}_{{m}}\left({\overline{\boldsymbol{s}}}_{{m}}\right) ||}^{2}}{2{a}_{m}^{2}}-\right.\\ \left.2{z}_{m}{Q}_{m}^{2}-\frac{{z}_{m}}{2}+{\dot{\omega }}_{m}\right) $ | (28) |
将式(24)、式(27)和式(28)代入式(26) ,可得
$ \begin{split} & {\dot{V}}_{m}\leqslant \left({{\varGamma }}_{1}\beta N\left(\psi \right) +1\right) \dot{\psi }-{\sum }_{i=1}^{m} {c}_{i}{z}_{i}^{2}-{c}_{1}{\xi }_{1}^{2}+ \\&\qquad \frac{1}{2}{\sum }_{i=1}^{m} \left({a}_{i}^{2}+{\lambda }_{i+1}^{2}+{\overline{\varepsilon }}_{i}^{2}\right) + \frac{1}{{r}_{0}}\tilde{\theta }\Bigg(\frac{{r}_{0}{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{||{\boldsymbol{S}}_{{1}}\left({x}_{1}\right) ||}^{2}}{2{a}_{1}^{2}}+\Bigg.\\&\qquad \Bigg.{\sum }_{i=2}^{m} \frac{{r}_{0}{z}_{i}^{2}{Q}_{i}^{2}{||{\boldsymbol{S}}_{{i}}\left({\overline{\boldsymbol{s}}}_{{i}}\right) ||}^{2}}{2{a}_{i}^{2}}-\dot{\hat{\theta }}\Bigg) + \Bigg(\frac{1}{2} + {\sum }_{i=2}^{m} \frac{{l}_{i}^{2}}{2}\Bigg) \parallel \boldsymbol{e}{\parallel }^{2} + \frac{{z}_{m+1}^{2}}{2} \end{split} $ | (29) |
Step n:设计第n步的Lyapunov函数
$ {V}_{n}={V}_{n-1}+\frac{{z}_{n}^{2}}{2} $ | (30) |
$ \begin{split} {\dot{V}}_{n}=&{\dot{V}}_{n-1}+{z}_{n}{Q}_{n}\left(u+{l}_{n}{e}_{1}\right) -{z}_{n}{\dot{\omega }}_{n}\leqslant \\& {\dot{V}}_{n-1}+\frac{{z}_{n}^{2}{Q}_{n}^{2}}{2}+\frac{{l}_{n}^{2}}{2}\parallel \boldsymbol{e}{\parallel }^{2}- {z}_{n}{\dot{\omega }}_{n}+{z}_{n}{Q}_{n}u \end{split} $ | (31) |
控制器
$ u=\frac{1}{{Q}_{n}}\left(-{c}_{n}{z}_{n}-\frac{{z}_{n}{Q}_{n}^{2}}{2}-\frac{{z}_{n}}{2}+{\dot{\omega }}_{n}\right) $ | (32) |
$ \dot{\hat{\theta }}={\sum }_{i=2}^{n-1} \frac{{r}_{0}{z}_{i}^{2}{Q}_{i}^{2}{||{\boldsymbol{S}}_{{i}}\left({\overline{\boldsymbol{s}}}_{{i}}\right) ||}^{2}}{2{a}_{i}^{2}}-\sigma \hat{\theta }+ \frac{{r}_{0}{\xi }_{1}^{2}{{\varGamma }}_{1}^{2}{||{\boldsymbol{S}}_{{1}}\left({x}_{1}\right) ||}^{2}}{2{a}_{1}^{2}}$ | (33) |
将式(29)、式(32)和式(33) 代入式(31) ,可得
$\begin{split} & {\dot{V}}_{n}\leqslant \left({{\varGamma }}_{1}\beta N\left(\psi \right) +1\right) \dot{\psi }-{c}_{1}{\xi }_{1}^{2}-{\sum }_{i=2}^{n}{c}_{i}{z}_{i}^{2}+\\ &\qquad \frac{1}{2}{\sum }_{i=1}^{n}({a}_{i}^{2}+{\overline{\varepsilon }}_{i}^{2}) +\frac{1}{2}{\sum }_{i=2}^{n} {\lambda }_{i}^{2}+\\ &\qquad \Bigg(\frac{1}{2}+{\sum }_{i=2}^{n} \frac{{l}_{i}^{2}}{2}\Bigg) \parallel \boldsymbol{e}{\parallel }^{2}+\frac{\sigma }{{r}_{0}}\tilde{\theta }\hat{\theta } \end{split} $ | (34) |
基于
$ V={V}_{{\boldsymbol{e}}}+{V}_{n}+\frac{1}{2}{\sum }_{i=2}^{n} {\lambda }_{i}^{2} $ | (35) |
对
$ \dot{V}={\dot{V}}_{e}+{\dot{V}}_{n}+{\sum }_{i=2}^{n} {\lambda }_{i}\left(-\frac{{\lambda }_{i}}{{\tau }_{i}}+{M}_{i}(\cdot ) \right) $ | (36) |
根据式(34)和式(36) ,使用Young’s不等式,可得
$\frac{\sigma }{{r}_{0}}\tilde{\theta }\hat{\theta }\leqslant \frac{\sigma }{2{r}_{0}}\left({\theta }^{*2}-{\tilde{\theta }}^{2}\right) $ | (37) |
$ {\sum}_{i=2}^{n} {\lambda }_{i}{M}_{i}(\cdot ) \leqslant \sum _{i=2}^{n} \left(\frac{{\lambda }_{i}^{2}{M}_{i}^{2}\left(\cdot \right) }{2}+\frac{1}{2}\right) $ | (38) |
将式(11) 、式(34) 、式(37) 和式(38) 代入式(36) ,整理后得
$\begin{split} & \dot{V}\leqslant -\Bigg(\nu -2\eta -\frac{1}{\varrho }-\frac{1}{2}-{\sum }_{i=2}^{n} \frac{{l}_{i}^{2}}{2}\Bigg) {\boldsymbol{e}}^{\text{T}}\boldsymbol{e}+\\&\qquad {\eta }^{-1}\parallel \boldsymbol{P}{\parallel }^{2}{\theta }^{*}+\varrho \parallel \boldsymbol{P}{\parallel }^{2}{\overline{\varpi }}^{2}+{\eta }^{-1}{\overline{\varepsilon }}_{0}^{2}+\\&\qquad \left({{\varGamma }}_{1}\beta N\left(\psi \right) +1\right) \dot{\psi }+\frac{1}{2}{\sum }_{i=1}^{n}({a}_{i}^{2}+{\overline{\varepsilon }}_{i}^{2}) -\\&\qquad {c}_{1}{\xi }_{1}^{2}+{\sum }_{i=2}^{n} \left(-\frac{1}{{\tau }_{i}}+\frac{1}{2}+\frac{{M}_{i}^{2}(\cdot ) }{2}\right) {\lambda }_{i}^{2}-\\&\qquad {\sum }_{i=2}^{n} {c}_{i}{z}_{i}^{2}-\frac{\sigma }{2{r}_{0}}{\tilde{\theta }}^{2}+\frac{\sigma }{2{r}_{0}}{\theta }^{*2}+\frac{n-1}{2} \end{split} $ | (39) |
定义以下参数
$ \begin{split} & q=\mathrm{m}\mathrm{i}\mathrm{n}\left\{2{c}_{i},\sigma ,\frac{2}{{\tau }_{i}}-1-{M}_{i}^{2}\left(\cdot \right) \right\} \\& \tilde{b}={\eta }^{-1}\parallel \boldsymbol{P}{\parallel }^{2}{\theta }^{*}+\varrho \parallel \boldsymbol{P}{\parallel }^{2}{\overline{\varpi }}^{2}+\\&\qquad \frac{1}{2}{\sum }_{i=1}^{n}({a}_{i}^{2}+{\overline{\varepsilon }}_{i}^{2}) +\frac{\sigma }{2{r}_{0}}{\theta }^{*2}+ \frac{n-1}{2}+{\eta }^{-1}{\overline{\varepsilon }}_{0}^{2} \end{split} $ |
存在参数
$ \nu > 2\eta +\frac{1}{\varrho }+\frac{1}{2}+{\sum }_{i=2}^{n} \frac{{l}_{i}^{2}}{2}+q{\lambda }_{\text{max}}\left(\boldsymbol{P}\right) $ |
式中:
$ \dot{V}\left(t\right) \leqslant -qV\left(t\right) +\left({{\varGamma }}_{1}\beta N\left(\psi \right) +1\right) \dot{\psi }+\overline{b} $ | (40) |
由引理2可知
$ V\left(t\right) \leqslant \frac{D}{q}+\left[V\left(0\right) -\frac{D}{q}\right]{{\rm{e}}}^{-qt} $ | (41) |
上式说明所有信号都是半全局一致最终有界。定义
$ \left|{\xi }_{1}\right|\leqslant \sqrt{\frac{2D}{q}+2\left[V\left(0\right) -\frac{D}{q}\right]{{\rm{e}}}^{-qt}} $ | (42) |
$ \left| {e_i} \right| \leqslant \parallel {\boldsymbol{e}}\parallel \leqslant \sqrt {\frac{{\dfrac{D}{q} + \Bigg( {V\left( 0 \right) - \dfrac{D}{q}} \Bigg){{\rm{e}}^{ - qt}}}}{{{\lambda _{{\rm{min}}}}\left( {{\boldsymbol{P}}} \right) }}}$ | (43) |
定义
定义
$\begin{split} & \frac{{\tilde{x}}_{1}^{2}}{{\mu }^{2}-{\tilde{x}}_{1}^{2}}={\xi }_{1}^{2}\leqslant \frac{2D}{q}+2\Bigg(V\left(0\right) -\frac{D}{q}\Bigg){{\rm{e}}}^{-qt}\\& {\tilde{x}}_{1}^{2}\leqslant \frac{{\phi }^{2}{\mu }^{2}}{1+{\phi }^{2}}\\& \left|{\tilde{x}}_{1}\right|\leqslant \sqrt{\frac{{\phi }^{2}}{1+{\phi }^{2}}}\left|\mu \right| < \left|\mu \right| \end{split} $ | (44) |
上式说明,通过指定性能控制设计,不仅能使系统跟踪误差始终被约束在指定性能函数的范围内,并且正确的参数选择能使系统跟踪误差变得足够小。
注4 从式(41) 可知,在基于反步法的框架下,根据构造的Lyapunov函数,所得控制律
综上所述,对原系统(1),利用设计的虚拟控制律(23)、(28) 和控制律(32),可以使原系统所有状态有界,系统跟踪误差满足设定的性能要求,并且能够在系统发生执行器故障的情况下,仍然保持良好的暂态性能和稳态性能。
3 仿真分析考虑如下纯反馈非线性系统
$ \left\{\begin{array}{l}{\dot{x}}_{1}={x}_{2}+0.5{x}_{2}{\mathrm{s}\mathrm{i}\mathrm{n}}^{2}{x}_{1}\\ {\dot{x}}_{2}={u}_{{\rm{f}}}-0.5{x}_{1}{x}_{2}^{3}\\ y={x}_{1}\end{array}\right. $ |
系统跟踪轨迹为
考虑系统执行器故障
$ {u}_{{\rm{f}}}=b(1-s) u+\varpi $ |
式中:参数设计为
本文仿真结果如图1~图7所示。图1说明在该控制方法下,系统输出对于目标函数有着良好的跟踪效果,并且系统状态
本文针对一类具有执行器故障的状态不可测纯反馈系统,考虑全状态约束和指定性能的约束条件,提出了一种自适应神经网络输出反馈容错控制方法。首先,建立状态观测器处理了系统状态不可测的问题。通过引入非线性映射,将状态约束系统转化为一个没有约束的新系统,从而解决了系统的全状态约束问题。其次,利用径向基神经网络逼近了系统中的未知非线性函数。此外,本文采用新的性能函数,有效地控制跟踪误差的收敛速度,并保证误差在所设置的时间内收敛。最后,仿真结果验证了本文控制算法的有效性。
[1] |
LI H Y, WU Y, CHEN M. Adaptive fault-tolerant tracking control fordiscrete-time multiagent systems via reinforcement learning algorithm[J].
IEEE Transactions on Cybernetics, 2021, 51(3): 1163-1174.
DOI: 10.1109/TCYB.2020.2982168. |
[2] |
田为刚, 王银河, 李玉姣. 一类非线性不确定系统的输出跟踪控制[J].
广东工业大学学报, 2015, 32(1): 91-97.
TIAN W G, WANG Y H, LI Y J. Fuzzy adaptive output tracking control for a class of uncertain nonlinear systems[J]. Journal of Guangdong University of Technology, 2015, 32(1): 91-97. DOI: 10.3969/j.issn.1007-7162.2015.01.019. |
[3] |
LIN G H, LI H Y, MA H, et al. Human-in-the-loop consensus control for nonlinear multi-agent systems with actuator faults[J].
IEEE/CAA Journal of Automatica Sinica, 2022, 9(1): 111-122.
|
[4] |
MA H, LI H Y, LIANG H J, et al. Adaptive fuzzy event-triggeredcontrol for stochastic nonlinear systems with full state constraints andactuator faults[J].
IEEE Transactions on Fuzzy Systems, 2019, 27(11): 2242-2254.
DOI: 10.1109/TFUZZ.2019.2896843. |
[5] |
郑晓宏, 董国伟, 周琪, 等. 带有输出约束条件的随机多智能体系统容错控制[J].
控制理论与应用, 2020, 37(5): 961-968.
ZHENG X H, DONG G W, ZHOU Q, et al. Fault-tolerantcontrol for stochastic multi-agent systems with output constraints[J]. Control Theory & Applications, 2020, 37(5): 961-968. |
[6] |
REN H R, KARIMI H R, LU R Q, et al. Synchronization of network systems via aperiodic sampled-data control with constant delay andapplication to unmanned ground vehicles[J].
IEEE Transactions on Industrial Electronics, 2020, 67(6): 4980-4990.
DOI: 10.1109/TIE.2019.2928241. |
[7] |
周琪, 陈广登, 鲁仁全, 等. 基于干扰观测器的输入饱和多智能体系统事件触发控制[J].
中国科学:信息科学, 2019, 49: 1502-1516.
ZHOU Q, CHEN G D, LU R Q, et al. Disturbance-observer-based event-triggered control for multi-agent systems withinput saturation[J]. Science China Information Sciences, 2019, 49: 1502-1516. |
[8] |
MA H, LI H Y, LU R Q, et al. Adaptive event-triggered control for aclass of nonlinear systems with periodic disturbances[J].
Science ChinaInformation Sciences, 2020, 63(5): 150212.
DOI: 10.1007/s11432-019-2680-1. |
[9] |
曾光, 孙炳达, 梁慧冰. 一种基于神经网络的直接自适应控制器[J].
广东工业大学学报, 2000, 17(1): 20-23.
ZENG G, SUN B D, LIANG H B. A direct adaptive controller based on neural network[J]. Journal of Guangdong University of Technology, 2000, 17(1): 20-23. DOI: 10.3969/j.issn.1007-7162.2000.01.005. |
[10] |
ZHOU Q, ZHAO S Y, LI H Y, et al. Adaptive neural network tracking control for robotic manipulators with dead zone[J].
IEEE Transactions on Neural Networks and Learning Systems, 2019, 30(12): 3611-3620.
DOI: 10.1109/TNNLS.2018.2869375. |
[11] |
LI Y M, TONG S C. Adaptive neural networks prescribed performance control design for switched interconnected uncertain nonlinear systems[J].
IEEE Transactions on Neural Networks and Learning Systems, 2018, 29(7): 3059-3068.
|
[12] |
REN B B, GE S S, TEE K P, et al. Adaptive neural control for outputfeedback nonlinear systems using a barrier Lyapunov function[J].
IEEE Transactions on Neural Networks, 2010, 21(8): 1339-1345.
DOI: 10.1109/TNN.2010.2047115. |
[13] |
ZHANG J X, YANG G H. Prescribed performance fault-tolerant control of uncertain nonlinear systems with unknown control directions[J].
IEEE Transactions on Automatic Control, 2017, 62(12): 6529-6535.
DOI: 10.1109/TAC.2017.2705033. |
[14] |
WANG C C, YANG G H. Observer-based adaptive prescribed performance tracking control for nonlinear systems with unknown controldirection and input saturation[J].
Neurocomputing, 2018, 284: 17-26.
DOI: 10.1016/j.neucom.2018.01.023. |
[15] |
KOSTARIGKA A K, ROVITHAKIS G A. Adaptive dynamic output feedback neural network control of uncertain MIMO nonlinearsystems with prescribed performance[J].
IEEE Transactions on Neural Networks and Learning Systems, 2012, 23(1): 138-149.
DOI: 10.1109/TNNLS.2011.2178448. |
[16] |
NI J K, AHN C K, LIU L, et al. Prescribed performance fixed-time recurrent neural network control for uncertain nonlinear systems[J].
Neurocomputing, 2019, 363: 351-365.
DOI: 10.1016/j.neucom.2019.07.053. |
[17] |
杨彬, 周琪, 曹亮, 等. 具有指定性能和全状态约束的多智能体系统事件触发控制[J].
自动化学报, 2019, 45(8): 1527-1535.
YANG B, ZHOU Q, CAO L, et al. Event-triggered control formulti-agent systems with prescribed performance and full state constraints[J]. Acta Automatica Sinica, 2019, 45(8): 1527-1535. |
[18] |
赵广磊, 高儒帅, 陈健楠. 具有执行器故障的四旋翼无人机自适应预定性能控制[J].
控制与决策, 2021, 36(9): 2103-2112.
ZHAO G L, GAO R S, CHEN J N. Adaptive prescribedperformance control of quadrotor with unknown actuator fault[J]. Control and Decision, 2021, 36(9): 2103-2112. DOI: 10.13195/j.kzyjc.2020.0083. |
[19] |
WANG F, CHEN B, LIN C, et al. Adaptive neural network finite-time output feedback control of quantized nonlinear systems[J].
IEEE Transactions on Cybernetics, 2018, 48(6): 1839-1848.
|
[20] |
WANG H H, CHEN B, LIN C, et al. Observer-based adaptive neuralcontrol for a class of nonlinear pure-feedback systems[J].
Neurocomputing, 2016, 171: 1517-1523.
DOI: 10.1016/j.neucom.2015.07.103. |
[21] |
ZHOU Q, SHI P, XU S Y, et al. Observer-based adaptive neural network control for nonlinear stochastic systems with time delay[J].
IEEE Transactionson Neural Networks and Learning Systems, 2013, 24(1): 71-80.
DOI: 10.1109/TNNLS.2012.2223824. |
[22] |
YANG H J, SHI P, ZHAO X D, et al. Adaptive output-feedback neuraltracking control for a class of nonstrict-feedback nonlinear systems[J].
Information Sciences, 2016, 334: 205-218.
|
[23] |
AREFIM M, ZAREI J, KARIMI H R. Adaptiveoutput feedback neural network control of uncertain non-affine systems with unknowncontrol direction[J].
Journal of the Franklin Institute, 2014, 351: 4302-4316.
DOI: 10.1016/j.jfranklin.2014.05.006. |
[24] |
毛骏, 张天平, 夏晓南, 等. 执行器有故障的多输入单输出系统的自适应输出反馈控制[J].
控制理论与应用, 2016, 33(4): 512-522.
MAO J, ZHANG T P, XIA X N, et al. Adaptive outputfeedback control for multi-input single-output systems with actuatorfailures[J]. Control Theory & Applications, 2016, 33(4): 512-522. DOI: 10.7641/CTA.2016.50037. |
[25] |
LI Y M, TONG S C. Adaptive neural networks decentralized FTC design for nonstrict-feedback nonlinear interconnected large-scale systems against actuator faults[J].
IEEE Transactions on Neural Networksand Learning Systems, 2017, 28(11): 2541-2554.
DOI: 10.1109/TNNLS.2016.2598580. |
[26] |
LI D P, LIU Y J, TONG S C, et al. Neural networks-based adaptivecontrol for nonlinear state constrained systems with input delay[J].
IEEE Transactions on Cybernetics, 2019, 49(4): 1249-1258.
|
[27] |
BOULKROUNE A. A fuzzy adaptive control approach for nonlinear systems with unknown control gain sign[J].
Neurocomputing, 2016, 179: 318-325.
DOI: 10.1016/j.neucom.2015.12.010. |
[28] |
ZHANG T P, XIA M Z, YI Y. Adaptive neural dynamic surface control of strict-feedback nonlinear systems with full state constraintsand unmodeled dynamics[J].
Automatica, 2017, 81: 232-239.
DOI: 10.1016/j.automatica.2017.03.033. |
[29] |
LIU C G, WANG H Q, LIU X P, et al. Adaptive finite-time fuzzy funnel control for nonaffine nonlinear systems[J].
IEEE Transactions on Systems, Man, And Cybernetics:Systems, 2021, 51(5): 2894-2903.
DOI: 10.1109/TSMC.2019.2917547. |