基于模糊学习观测器的一类具有时变时滞的Takagi-Sugeno模糊系统的鲁棒故障估计
Robust Fault Estimation for a Class of Takagi-Sugeno Fuzzy Systems with Time-Varying Delay via Fuzzy Learning Observer
摘要:本文研究一类具有连续状态时变时滞、执行器故障和范数有界外部干扰的Takagi-Sugeno(T-S)模糊系统的鲁棒故障估计问题。利用H∞优化技术,构造了一种新颖的模糊学习观测器,实现了系统状态和执行器故障的同时估计。基于Lyapunov稳定性分析方法,以一组线性矩阵不等式(LMIs)的解形式给出了误差动态系统的稳定性分析和一个保守性更小的时滞相关的充分条件。最后,通过一个动态模型的仿真结果说明了所提方法的有效性。
Abstract:This paper concerns the problem of robust fault estimation for a class of Takagi-Sugeno (T-S) fuzzy systems subject to continuous state time-varying delay, actuator faults and norm-bound external disturbances. A novel fuzzy learning observer is constructed to achieve simultaneous estimation of system states and actuator faults by using theH∞optimization technique. Based on the Lyapunov method, stability analysis for the error dynamic and one less conservative delay dependent sufficient conditions are formulated in terms of solutions in a set of Liner Matrix Inequalities (LMIs). Finally, simulation results of a dynamic model are illustrated to show the effectiveness of the proposed approaches.
文章引用:盛光玉, 刘姿君, 葛春婷, 孙超. 基于模糊学习观测器的一类具有时变时滞的Takagi-Sugeno模糊系统的鲁棒故障估计[J]. 动力系统与控制, 2024, 13(3): 91-104. https://doi.org/10.12677/dsc.2024.133009

1. 引言

现代工程实际系统复杂程度的增加会相应地增加故障系统崩溃的可能性,如传感器故障、执行器故障和过程故障,甚至元件故障等。为了满足工程系统对可靠性和安全性的严格要求,近几十年来,故障检测与隔离(FDI)、容错控制(FTC)和故障估计在学术研究和实际应用中都受到了相当大的关注,并且已经建立了相关技术和工具(见[1]-[4])。FDI技术用于检测系统是否发生故障以及故障发生在何处,但在实际系统中很难准确提供故障的大小和形状。然而,故障估计可以获得故障的大小、形状和持续时间[5][6]的精确信息。通过故障估计得到的估计信息用来重构故障信号,以保证系统的稳定性和可接受的性能。

时滞现象存在于各种工程、生物、经济、力学、物理和医疗系统中,如动力系统中的长输电线、轧机、飞机、核反应堆、神经网络和船舶稳定性等。时滞的存在会导致系统的不稳定,降低系统性能。因此,对具有时滞的系统进行研究是有必要的。

故障估计首次被应用于线性系统中[7][8],但在现实中,绝大多数工程系统都包含非线性行为。众所周知,T-S模糊模型是利用一组“IF-THEN”规则来逼近一系列高度复杂的非线性动态系统的好方法[9]。T-S模糊模型是由若干个局部线性子系统表示的一类非线性系统,它的一个基本优点是可以简单有效地表示一个非线性系统,人们对此进行了大量重要的研究[10]-[12]。在文献[5]中,研究了一类在外部干扰下,同时存在传感器和执行器故障的Takagi-Sugeno (T-S)模糊系统的故障估计问题。在后者中,首先构造了一个学习观测器来估计系统状态和执行器故障,然后对一类T-S模糊系统进行了时滞分析。在文献[13]中,作者考虑了非线性离散时滞系统的T-S模糊模型,设计了模糊故障估计观测器,实现了对系统状态、传感器和执行器故障的同时估计。遗憾的是,这些方法没有考虑连续时变时滞。文献[14]考虑了一类带有未知有界干扰的T-S模糊模型的非线性系统,建立了一个自适应观测器来同时估计系统状态和执行器故障。此外,它可以提高系统的性能,但它没有考虑时滞。针对T-S模糊时滞系统[15],作者引入了一种新颖的最小范数最小二乘解进行故障估计补偿。文献[16]是关于故障估计的现有工作,其中作者同时考虑了故障和未知输入,然后同时考虑模糊未知输入观测器和模糊滑模观测器。针对存在故障和传感器扰动的T-S模糊系统[17],提出一种改进的自适应观测器来实现系统状态和故障参数的同时估计。文献[18]研究了离散时间T-S模糊非线性系统的故障估计问题,提出了一种新的故障观测器。但是这些文献都没有考虑时滞的影响。文献[19]研究了一类T-S仿射系统的事件触发 H 状态估计问题,提出了一种基于分段Lyapunov-Krasovskii函数泛函和Finsler引理的事件触发状态观测器。然而,时变时滞和执行器故障尚未讨论。

据我们所知,目前针对具有时变时滞、执行器故障和外部扰动的T-S模糊系统的鲁棒故障估计的研究成果较少,这激发了本文的研究动机。本文设计了一种模糊学习观测器来同时估计系统状态和执行器故障,同时也考虑了系统状态的时变时滞。然后利用 H 优化技术,证明了模糊学习观测器设计的可行性,并使用MATLAB软件的LMI工具箱[20]来求解一组LMIs来计算它们的增益。本文的主要贡献概括如下:

1) 针对具有时变时滞、执行器故障和外部干扰的T-S模糊模型的连续时间系统,本文提出了一种模糊学习观测器,可以实现系统状态和执行器故障的同时估计。

2) 与传统的故障估计方法不同,该方法利用前一次迭代中的输出估计误差和故障估计信息,在当前迭代中进行故障估计,以改善估计结果。

3) 对误差系统建立了新的Lyapunov函数,并引入更多的可调参数以降低系统的保守性。

本文的内容安排如下:第2节概述了具有时变时滞的T-S模糊系统的问题描述,并给出了一些预备知识。第3节给出了主要的理论结果,设计了模糊学习观测器来同时估计系统状态和执行器故障,然后提出李雅普诺夫函数,得到期望的故障估计结果,并利用LMI进行求解。第4节给出了一个仿真例子来说明所提方法的有效性。第5节概括了一些结论性的贡献。

在整篇文章中,一些必要的记号被考虑如下: A < 0 ( A > 0 )表示A的负(正)定矩阵; λ max λ min 分别是A最大值和最小值的表达式, s y m ( A ) 表示 A + A T ;“ ”表示“所有或任意”;“ * ”表示对称矩阵中对称位置相同的值, 表示欧式空间;I表示单位矩阵。

2. 问题描述

本文所考虑的具有时变和执行器故障的连续时间T-S模糊系统是由一类if-then规则组成的,其T-S模糊模型的第i条规则由以下方程描述:

i条规则:如果 θ 1 μ i 1 ,…, θ p μ i p ,那么

{ x ˙ ( t ) = A i x ( t ) + A h i x ( t h ( t ) ) + B i u ( t ) + E i f a ( t ) + D i d ( t ) y ( t ) = C x ( t ) (1)

其中, θ ( x ( t ) ) = [ θ 1 ( x ( t ) ) , θ 2 ( x ( t ) ) , , θ p ( x ( t ) ) ] ,表示系统可测量的前提变量。模糊集 μ i j ( i = 1 , , r ; j = 1 , , p ) 用表示隶属度函数;rp分别表示if-then规则和前提变量的个数, x ( t ) n 分别是系统状态向量, u ( t ) m 是控制输入, y ( t ) 表示输出, f a ( t ) p 表示执行器故障, d ( t ) v 表示未知的外部干扰输入或属于的不确定性,它属于 L 2 [ 0 , ] 。它们可以是常数,也可以是时变的。 A i A h i B i E i D i C是具有适当维数已知的实常数矩阵。常数矩阵 E i 为满列秩,即 r a n k ( E ) = r h ( t ) 为时变时滞。

通过模糊融合,得出整体模糊系统模型如下:

{ x ˙ ( t ) = i = 1 r h i ( θ ( x ( t ) ) ) × [ A i x ( t ) + A h i x ( t h ( t ) ) + B i u ( t ) + E i f a ( t ) + D i d ( t ) ] y ( t ) = C x ( t ) (2)

其中

h i ( θ ( x ( t ) ) ) = v i ( θ ( x ( t ) ) ) i = 1 r v i ( θ ( x ( t ) ) ) v i ( θ ( x ( t ) ) ) = j = 1 p μ i j ( θ i ( x ( t ) ) )

其中, μ i j ( θ i ( x ( t ) ) ) 表示 θ i ( x ( t ) ) μ i j 中的隶属度。显然, 0 h i ( θ ( x ( t ) ) ) 1 i = 1 r h i ( θ ( x ( t ) ) ) = 1 。在接下来的内容中,我们简单地用 h i 表示 h i ( θ ( x ( t ) ) )

为了计算的方便,系统(2)可以改写为

{ x ˙ ( t ) = A ( t ) x ( t ) + A h ( t ) x ( t h ( t ) ) + B ( t ) u ( t ) + E ( t ) f a ( t ) + D ( t ) d ( t ) y ( t ) = C x ( t ) (3)

其中 A ( t ) = i = 1 r h i A i B ( t ) = i = 1 r h i B i E ( t ) = i = 1 r h i E i D ( t ) = i = 1 r h i D i A h ( t ) = i = 1 r h i A h i

为了得到主要结果,需要下面的假设和引理。

假设1:假设 ( A i , B i ) 是局部能控的, ( A i , C ) 是局部能观的。

假设2[21]:时滞 h ( t ) 为时变连续函数,满足 0 h ( t ) h h ˙ ( t ) h m ,且h h m 为常数。

假设3: r a n k ( C E ) = r

假设4: r a n k [ A i s I E i C 0 ] = n + r a n k ( E ) s Re ( s ) 0 i = [ 1 , , r ]

假设5: ( A i , E i , C ) 的不变零点存在于开左半平面。

假设6[22]:我们假设 f ˙ a ( t ) f ˙ a ( t ) 是有界的, f a ( t ) 对时间t的导数的范数满足 f ˙ a ( t ) f 1 a f ˙ a ( t ) f 1 c ,其中 0 < f 1 a < 0 < f 1 c <

假设7[23]:当初始条件为零时, V ( x ^ , t ) | t = 0 = 0 V ( x ^ , t ) | t = 0

引理1[24]:对于给定的两个近似维数的矩阵XY,存在某个正定对称矩阵P,使得下列不等式成立:

2 X T Y X T P X + Y T P Y (4)

引理2[25]:对任一给定的正定矩阵 M n × n ,定义下列满足 β α 的标量 α > 0 β > 0 及向量函数 x ( s ) 的积分项,则下列积分不等式成立, [ α , β ] R n

( α β x ( s ) d s ) T M ( α β x ( s ) d s ) ( β α ) ( α β x ( s ) M x ( s ) d s ) (5)

3. 模糊学习观测器的设计及稳定性分析

3.1. 模糊学习观测器的设计

在这一部分,为了实现系统(1)的系统状态和执行器故障的同时估计,提出了如下的模糊学习观测器:

{ x ˙ ( t ) = A ( t ) x ^ ( t ) + A h ( t ) x ^ ( t h ( t ) ) + B ( t ) u ( t ) + E ( t ) f ^ a ( t ) + L ( t ) ( y ( t ) y ^ ( t ) ) y ^ ( t ) = C x ^ ( t ) f ^ a ( t ) = K 1 f a ( t d ) + K 2 ( t ) ( y ( t ) y ^ ( t ) ) (6)

其中 x ^ ( t ) n 是状态的估计, y ^ ( t ) p 表示估计的输出向量。 f ^ a ( t ) m 表示执行器故障的估计,它与采样时刻 t d 的先前信息和当前的输出误差都有关。参数d表示学习间隔。对角矩阵 K 1 = d i a g { σ 1 σ m } 其中 σ i ( 0 , 1 ] K 1 K 2 ( t ) L ( t ) 是待设计的近似维数增益矩阵,其中

K 2 ( t ) = i = 1 r h i K 2 i L ( t ) = i = 1 r h i L i

让我们定义 e x ( t ) = x ¯ ( t ) x ¯ ^ ( t ) e f ( t ) = f a ( t ) f ^ a ( t ) e y ( t ) = y ( t ) y ^ ( t ) 。则由式(3)和式(6)可得动态系统误差为

{ e ˙ x ( t ) = ( A ( t ) L ( t ) C ) e x ( t ) + A h ( t ) e x ( t h ( t ) ) + E ( t ) e f ( t ) e y ( t ) = C e x ( t ) e f ( t ) = K 1 e f ( t d ) + K 2 ( t ) C e x ( t ) + f ˜ a ( t ) (7)

其中

f ˜ a ( t ) = f a ( t ) K 1 f a ( t d )

为了简单起见,我们给出下面的向量:

ξ T ( t ) = [ e x T ( t ) e x T ( t h ( t ) ) e x T ( t h ) d T ( t ) ]

Γ 1 ( t ) = [ A ( t ) L ( t ) C A h ( t ) 0 D ( t ) ] Γ 2 ( t ) = [ A ( t ) L ( t ) C 0 0 0 ]

Γ 3 ( t ) = [ 0 A h ( t ) 0 0 ] Γ 4 ( t ) = [ 0 0 0 D ( t ) ] Γ 5 ( t ) = [ E ( t ) K 2 ( t ) C 0 0 0 ]

3.2. 模糊学习观测器的稳定性分析

定理1在假设1~7下考虑系统(3)。给定一个实标量,在以下 H 性能下,学习模糊观测器能够保证时变动态的渐近稳定性,即

0 e y ( s ) 2 d s 0 d ( s ) 2 d s (8)

如果存在实标量h h m λ 1 δ 和对称正定矩阵 P > 0 Q 1 > 0 Q 2 > 0 R > 0 W > 0 Y ( t ) ,和合适的矩阵 K 1 K 2 ( t ) ,使得(9)~(11)成立:

E ( t ) P = λ 1 K 2 ( t ) C (9)

Λ ( t ) = ( λ 1 + λ 1 η ) K 1 T K 1 W + 12 h 2 K 1 T E T ( t ) R E ( t ) K 1 0 (10)

ϒ ( t ) = [ ϒ ˜ ( t ) h Γ 1 T ( t ) h Γ 2 T ( t ) h Γ 3 T ( t ) h Γ 4 T ( t ) 12 h Γ 5 T ( t ) * 2 δ P + δ 2 R 0 0 0 0 * * 2 δ P + δ 2 R 0 0 0 * * * 2 δ P + δ 2 R 0 0 * * * * 2 δ P + δ 2 R 0 * * * * * 2 δ I + δ 2 R ] < 0 (11)

其中

ϒ ˜ ( t ) = [ ϒ ˜ 1 P A h ( t ) + R 0 P D ( t ) * ε ( 1 h m ) Q 1 + R R 0 * * Q 2 R 0 * * * γ 2 I ]

其中

ϒ ˜ 1 = s y m ( P A ( t ) Y ( t ) C ) + ε Q 1 + Q 2 + C T C R

然后,观测器增益矩阵可以相应地计算为

Y ( t ) = P L ( t ) (12)

1对于误差动态系统(7),我们可以知道新的矩阵 A ( t ) A h ( t ) E ( t ) 是已知的矩阵,并且矩阵 L ( t ) K 2 ( t ) 是必须设计的。当 d ( t ) = 0 时,所提出的模糊学习观测器能够保证带有时变状态时滞的误差动态系统(7)的渐近稳定性;否则,必须满足 H 性能(8)。

证明:在本文中,Lyapunov-Krasovskii函数泛函是在如下形式下构造的:

V ( t ) = V 1 ( t ) + V 2 ( t ) + V 3 ( t ) + V 4 ( t ) (13)

V 1 ( t ) = e x T ( t ) P e x ( t ) (14)

V 2 ( t ) = ε t h ( t ) t e x T ( s ) Q 1 e x ( s ) d s + t h t e x T ( s ) Q 2 e x ( s ) d s (15)

V 3 ( t ) = h 0 t + θ t e ˙ x T ( s ) R e ˙ x ( s ) d s d θ (16)

V 4 ( t ) = t h t e f T ( s ) W e f ( s ) d s (17)

那么,可以得到 V ( t ) 的时间导数

V ˙ ( t ) = V ˙ 1 ( t ) + V ˙ 2 ( t ) + V ˙ 3 ( t ) + V ˙ 4 ( t )

V ˙ 1 ( t ) = e x T ( t ) s y m ( P A ( t ) L ( t ) C ) e x ( t ) + 2 e x T ( t ) P A h ( t ) e x ( t h ( t ) ) + 2 e x T ( t ) P E ( t ) e f ( t ) + 2 e x T ( t ) P D ( t ) d ( t ) (18)

V ˙ 2 ( t ) = ε e x T ( t ) Q 1 e x ( t ) ε ( 1 h ˙ ( t ) ) e x T ( t h ( t ) ) Q 1 e x ( t h ( t ) ) + e x T ( t ) Q 2 e x ( t ) e x T ( t h ) Q 2 e x ( t h ) (19)

考虑假设2,我们有

V ˙ 2 ( t ) ε e x T ( t ) Q 1 e x ( t ) ε ( 1 h m ) e x T ( t h ( t ) ) Q 1 e x ( t h ( t ) ) + e x T ( t ) Q 2 e x ( t ) e x T ( t h ) Q 2 e x ( t h ) (20)

V ˙ 3 ( t ) = h 2 e ˙ x T ( t ) R e ˙ x ( t ) h t h t e ˙ x T ( t ) R e ˙ x ( t ) d s (21)

V ˙ 4 ( t ) = e f T ( t ) W e f ( t ) e f T ( t d ) W e f ( t d ) (22)

λ 1 = λ max ( W ) ,然后我们就能得到

e x T ( t ) P E ( t ) e f ( t ) + e f T ( t ) W e f ( t ) 2 e x T ( t ) P E ( t ) e f ( t ) + λ 1 e f T ( t ) e f ( t ) (23)

把(7)式代入到不等式(23)式中,得到

2 e x T ( t ) P E ( t ) e f ( t ) + λ 1 e f T ( t ) e f ( t ) = 2 e x T ( t ) ( P E ( t ) λ 1 C T ( t ) K 2 T ( t ) ) K 1 e f ( t d ) + 2 e x T ( t ) ( P E ( t ) λ 1 C T ( t ) K 2 T ( t ) ) K 2 ( t ) C e f ( t d ) + e x T ( t ) ( P E ( t ) λ 1 C T ( t ) K 2 T ( t ) ) f ˜ a ( t ) + λ 1 e f T ( t d ) K 1 T K 1 e f ( t d ) + 2 λ 1 e f T ( t d ) K 1 T f ˜ a ( t ) + λ 1 f ˜ a T ( t ) f ˜ a ( t ) (24)

根据引理1,

2 e f T ( t d ) K 1 T f ˜ a ( t ) η e f T ( t d ) K 1 T K 1 e f ( t d ) + 1 η f ˜ a T ( t ) f ˜ a ( t ) (25)

考虑(9)式和(25)式,得到

2 e x T ( t ) P E ( t ) e f ( t ) + λ 1 e f T ( t ) e f ( t ) ( λ 1 + λ 1 η ) e f T ( t d ) K 1 T K 1 e f ( t d ) + ( λ 1 + λ 1 η ) f ˜ a T ( t ) f ˜ a ( t ) (26)

接下来,

e ˙ x ( t ) R e ˙ x ( t ) = e x T ( t ) ( A ( t ) L ( t ) C ) T R ( A ( t ) L ( t ) C ) e x ( t ) + 2 e x T ( t ) ( A ( t ) L ( t ) C ) T R A h ( t ) e x ( t h ( t ) ) + 2 e x T ( t ) ( A ( t ) L ( t ) C ) T R E ( t ) e f ( t ) + e x T ( t ) ( A ( t ) L ( t ) C ) T R D ( t ) d ( t ) + e x T ( t h ( t ) ) A h T ( t ) R A h ( t ) e x ( t h ( t ) ) + 2 e x T ( t h ( t ) ) A h T ( t ) R E ( t ) e f ( t ) + 2 e x T ( t h ( t ) ) A h T ( t ) R D ( t ) d ( t ) + e f T ( t ) E T ( t ) R E ( t ) e f ( t ) + 2 e f T ( t ) E T ( t ) R D ( t ) d ( t ) + d T ( t ) D T ( t ) R D ( t ) d ( t ) (27)

正如我们所知,R是正定的,则

2 e x T ( t ) ( A ( t ) L ( t ) C ) T R E ( t ) e f ( t ) e x T ( t ) ( A ( t ) L ( t ) C ) T R ( A ( t ) L ( t ) C ) e x ( t ) + e f T ( t ) E T ( t ) R E ( t ) e f ( t ) (28)

2 e x T ( t h ( t ) ) A h T ( t ) R E ( t ) e f ( t ) e x T ( t h ( t ) ) A h T ( t ) R e x ( t h ( t ) ) + e f T ( t ) E T R E ( t ) e f ( t ) (29)

2 e f T ( t ) E T ( t ) R D ( t ) d ( t ) e f T ( t ) E T ( t ) R E ( t ) e f ( t ) + d T ( t ) D T ( t ) R D ( t ) d ( t ) (30)

将(7)式代入 e f T ( t ) E T ( t ) R E ( t ) e f ( t ) ,则

e f T ( t ) E T ( t ) R E ( t ) e f ( t ) e f T ( t d ) K 1 T E T ( t ) R E ( t ) K 1 e f ( t d ) 2 e f T ( t d ) K 1 T E T ( t ) R E ( t ) K 2 ( t ) C e x ( t ) + 2 e f T ( t d ) K 1 T E T ( t ) R E ( t ) f ˜ a ( t ) + e x ( t ) C T K 2 T E T ( t ) R K 2 T ( t ) C x ( t ) 2 e x T ( t ) C T K 2 T ( t ) E T ( t ) R E ( t ) f ˜ a T ( t ) + f ˜ a T ( t ) E T ( t ) R E ( t ) f ˜ a ( t ) (31)

2 e f T ( t d ) K 1 T E T ( t ) R E ( t ) K 2 ( t ) C e x ( t ) e f T ( t d ) K 1 T E T ( t ) R E ( t ) K 1 e f ( t d ) + e x T C T K 2 T ( t ) E T ( t ) R E ( t ) K 2 ( t ) C e x ( t ) (32)

2 e f T ( t d ) K 1 T E T ( t ) R E ( t ) f ˜ a ( t ) e f T ( t d ) K 1 T E T ( t ) R E ( t ) K 1 e f ( t d ) + f ˜ a T ( t ) E T ( t ) R E ( t ) f ˜ a ( t ) (33)

2 e x T ( t ) C T K 2 T ( t ) E T ( t ) R E ( t ) f ˜ a T ( t ) e x T ( t ) C T K 2 T ( t ) E T ( t ) R E ( t ) K 2 ( t ) e x ( t ) + f ˜ a T ( t ) E T ( t ) R E ( t ) f ˜ a ( t ) (34)

因此

e ˙ x T ( t ) R e ˙ x ( t ) 2 e x T ( t ) ( A ( t ) L ( t ) C ) T R ( A ( t ) L ( t ) C ) e x ( t ) + 2 T ( t ) ( A ( t ) L ( t ) C ) T R A h ( t ) e x ( t h ( t ) ) + 2 e x T ( t ) ( A ( t ) L ( t ) C ) T R D ( t ) d ( t ) + 2 e x T ( t h ( t ) ) A h T ( t ) R A h ( t ) e x ( t h ( t ) ) + 2 e x T ( t h ( t ) ) A h T ( t ) R D ( t ) d ( t ) + 2 d T ( t ) D T ( t ) R D ( t ) d ( t ) + 12 e x T ( t ) C T K 2 ( t ) E T ( t ) R E ( t ) K 2 T ( t ) C e x ( t ) + 12 e f T ( t d ) K 1 T E T ( t ) R E ( t ) K 1 e f ( t d ) + 12 f ˜ a T ( t ) E T ( t ) R E ( t ) f ˜ a ( t ) ξ T ( t ) ( Γ 1 T ( t ) R Γ 1 ( t ) + Γ 2 T ( t ) R Γ 2 ( t ) + Γ 3 T ( t ) R Γ 3 ( t ) + Γ 4 T ( t ) R Γ 4 ( t ) + 12 Γ 5 T ( t ) R Γ 5 ( t ) ) ξ ( t ) + 12 e f T ( t d ) K 1 T E T ( t ) R E ( t ) K 1 e f ( t d ) + 12 f ˜ a T ( t ) E T ( t ) R E ( t ) f ˜ a ( t ) (35)

根据引理2,我们得到

h t h t e ˙ x T ( t ) R e ˙ x T ( t ) e x T ( t ) R e x ( t ) + 2 e x T ( t ) R e x T ( t h ( t ) ) 2 e x T ( t h ( t ) ) R e x ( t h ( t ) ) + 2 e x T ( t h ( t ) ) R e x ( t h ) e x T ( t h ) R e x ( t h ) (36)

将(13),(17),(21),(30)和(31)代入到Lyapunov函数的导数中,得到

V ˙ ( t ) e x T ( t ) [ s y m ( A ( t ) L ( t ) C ) + ε Q 1 + Q 2 R ] e x ( t ) + 2 e x T ( t ) ( P A h ( t ) + R ) e x T ( t h ( t ) ) e x T ( t h ( t ) ) ( ε ( 1 h m ) Q 1 + 2 R ) e x ( t h ( t ) ) + 2 e x T ( t h ( t ) ) R e x ( t h ) e x T ( t h ) ( Q 2 + R ) e x ( t h ) + h 2 ξ T ( t ) ( Γ 1 T ( t ) R Γ 1 ( t ) + Γ 2 T ( t ) R Γ 2 ( t ) + Γ 3 T ( t ) R Γ 3 ( t ) + Γ 4 T ( t ) R Γ 4 ( t ) + 12 Γ 5 T ( t ) R Γ 5 ( t ) ξ ( t ) + e f T ( t d ) Λ e f ( t d ) + f ˜ a T ( t ) ( λ 1 + λ 1 η + 12 h 2 + λ 2 ) f ˜ a ( t ) (37)

给定 J ( t ) = V ˙ ( t ) + e y T ( t ) e y ( t ) γ 2 d T ( t ) d ( t )

然后,令 λ 2 = λ max ( E ( t ) R E ( t ) )

J ( t ) ξ T ( t ) ϒ ˜ ( t ) ξ T ( t ) + h 2 ξ T ( t ) ( Γ 1 T ( t ) R Γ 1 ( t ) + Γ 2 T ( t ) R Γ 2 ( t ) + Γ 3 T ( t ) R Γ 3 ( t ) + Γ 4 T ( t ) R Γ 4 ( t ) + 12 Γ 5 T ( t ) R Γ 5 ( t ) ) ξ ( t ) + e f T ( t d ) Λ e f ( t d ) + f ˜ a T ( t ) ( λ 1 + λ 1 η + 12 h 2 + λ 2 ) f ˜ a ( t ) (38)

我们知道,如果对于任意的标量 δ ,有 ( δ R P ) R 1 ( δ R P ) 0 ,那么 P R 1 P 2 δ P + δ 2 R 。通过Schur补,(33)可以写成如下形式:

J ( t ) ξ T ( t ) ϒ ( t ) ξ T ( t ) + e f T ( t d ) Λ e f ( t d ) + f ˜ a T ( t ) ( λ 1 + λ 1 η + 12 h 2 + λ 2 ) f ˜ a ( t ) (39)

如果条件(9)和条件(10)成立,则

J ( t ) ζ ξ ( t ) + ρ

其中,

ζ = λ min ( d i a g { ϒ ( t ) , Λ } ) ρ = ( λ 1 + λ 1 η + 12 h 2 λ 2 ) f 2

因此,当 ζ ξ ( t ) ρ J ( t ) 0 成立时,在Lyapunov稳定性理论下 ξ ( t ) 将收敛到一个小的集合

ψ = { ξ ( t ) | | ξ ( t ) | | 2 ξ ρ } ,从而 ξ ( t ) 是一致有界的,这表明系统(7)的渐近稳定性。证明完毕。

2:我们需要将定理1中的方程(8)转化为下面的低阶优化问题,通过MATLAB的LMI工具箱可以很容易地求解。

Min u > 0 , s.t.

[ u I E T ( t ) P λ 1 K 2 ( t ) C * u I ] > 0 (40)

定理2考虑系统(2),对于给定的实标量h h m λ 1 δ γ ,误差动态系统是渐近稳定的(其中 d ( t ) = 0 ),且满足给定的性能(8),如果存在对称正定矩阵 P > 0 Q 1 > 0 Q 2 > 0 R > 0 W > 0 Y i ( i = 1 , 2 , , r ) 和合适的矩阵 K 1 K 2 i ( i = 1 , 2 , , r ) ,使得下列条件成立:

E i P = λ 1 K 2 i C ( i = 1 , 2 , , r ) (41)

ϒ i i + Λ i 0 ( i = 1 , 2 , , r ) (42)

ϒ i j + ϒ j i + Λ i + Λ j 0 , 1 i j r (43)

其中

Λ i = ( λ 1 + λ 1 η ) K 1 T K 1 W + 12 h 2 K 1 T E i T R E i K 1 0

ϒ i j = [ ϒ ˜ i j h Γ 1 i T h Γ 2 i T h Γ 3 i T h Γ 4 j T 12 h Γ 5 j T * 2 δ P + δ 2 R 0 0 0 0 * * 2 δ P + δ 2 R 0 0 0 * * * 2 δ P + δ 2 R 0 0 * * * * 2 δ P + δ 2 R 0 * * * * * 2 δ I + δ 2 R ] 0

其中

Γ 1 i = [ A i L i C A h i 0 D i ] Γ 2 i = [ A i L i C 0 0 0 ]

Γ 3 i = [ 0 A h i 0 0 ] Γ 4 j = [ 0 0 0 D j ] Γ 5 j = [ E K 2 j C 0 0 0 ]

ϒ ˜ ( t ) = [ ϒ ˜ 1 i j P A h i + R 0 P D i * ε ( 1 h m ) Q 1 + R R 0 * * Q 2 R 0 * * * γ 2 I ]

ϒ ˜ 1 i j = s y m ( P A i Y i C ) + ε Q 1 + Q 2 + C T C R

定理2的证明与本文定理1类似。

4. 仿真实验结果

在该部分中,以下两个规则的T-S模糊模型被用来说明所提出方法的有效性。然后给出了整体的模糊系统为

{ x ˙ ( t ) = i = 1 2 h i × [ A i x ( t ) + A h i x ( t h ( t ) ) + B i u ( t ) + E i f a ( t ) + D i d ( t ) ] y ( t ) = C x ( t )

其中

A 1 = [ 10 1 2 48 2 0 1.5 1.5 30 ] A 2 = [ 10 1 2 48 2 0 1 1 20 ] A h 1 = [ 0.5 0 1 0.5 1 0.5 2 0 1 ] A h 2 = [ 0.5 0 1 0.5 1 0.5 0.25 0 0.5 ] B 1 = B 2 = [ 1 1 1 ] E 1 = E 2 = [ 1 1 1 ] D 1 = D 2 = [ 1 0 0 1 0 0 1 0 0 ] C = [ 0 0 1 ]

在仿真过程中,选取隶属度函数为 h 1 = 1 / ( 1 + exp ( 1 + x 1 ) ) h 2 = 1 h 1 ,并选取仿真初始状态为 x ( 0 ) = [ 0.1 0.1 0.2 ] T x ^ ( 0 ) = [ 0 0 0 ] T 。扰动 d ( t ) 为带限白噪声,其中功率为0.001,采样时间为0.1。我们选择W为单位矩阵。假设作用于系统状态的时变时滞函数为 h ( t ) = 0.3 + 0.3 sin ( t ) 。选取其余参数值为 h = 0.3 h m = 0.6 γ = 0.5 u = 10 5 η = 0.01 ε = 1.3 d = 0.004 δ = 30 K 1 = 0.995 。考虑如下的时变执行器故障为

f a 1 ( t ) = { 5 sin ( t ) , t 2 0 , other

利用MATLAB的LMI工具箱,可以方便地求解定理2中的(41)、(42)和(43)。模糊学习观测器的增益矩阵可以得到如下形式:

K 21 = K 22 = 11.0325 L 1 = [ 77.2912 36.4065 61.5652 ] L 2 = [ 57.1088 12.4299 48.7172 ]

系统状态和执行器故障及其估计趋势如图1~4所示,本文设计的模糊学习观测器对系统的状态和执行器故障都有很好的估计。基于时变时滞,我们可以能够总结即使存在执行器故障和外部干扰,模糊学习观测器也可以迅速恢复系统的性能和稳定性。

Figure 1. System state x 1 ( t ) and its estimation x ^ 1 ( t ) under time-varying actuator fault

图1. 时变执行器故障下系统状态 x 1 ( t ) 及它的估计 x ^ 1 ( t )

Figure 2. System state x 2 ( t ) and its estimation x ^ 2 ( t ) under time-varying actuator fault

图2. 时变执行器故障下系统状态 x 2 ( t ) 及它的估计 x ^ 2 ( t )

Figure 3. System state x 3 ( t ) and its estimation x ^ 3 ( t ) under time-varying actuator fault

图3. 时变执行器故障下系统状态 x 3 ( t ) 及它的估计 x ^ 3 ( t )

Figure 4. Actuator fault f a 1 ( t ) and its estimation f ^ a 1 ( t )

图4. 执行器故障 f a 1 ( t ) 及它的估计 f ^ a 1 ( t )

当执行器故障是以下的恒定故障时,我们取 K 1 = 0.995 d = 0.01 f a 2 ( t ) = { 5 , t 2 0 , other

系统状态和执行器故障及其估计趋势如图5~图7所示,由图8可知,执行器故障出现在2 s时刻。图8显示了故障估计 f ^ a 2 ( t ) 可以很好地跟踪故障 f a 2 ( t ) ,并且当状态为区间时变时,估计误差很小。

Figure 5. System state x 1 ( t ) and its estimation x ^ 1 ( t ) under time-varying actuator fault

图5. 时变执行器故障下系统状态 x 1 ( t ) 及它的估计 x ^ 1 ( t )

Figure 6. System state x 2 ( t ) and its estimation x ^ 2 ( t ) under time-varying actuator fault

图6. 时变执行器故障下系统状态 x 2 ( t ) 及它的估计 x ^ 2 ( t )

Figure 7. System state x 3 ( t ) and its estimation x ^ 3 ( t ) under time-varying actuator fault

图7. 时变执行器故障下系统状态 x 3 ( t ) 及它的估计 x ^ 3 ( t )

Figure 8. Actuator fault f a 2 ( t ) and its estimation f ^ a 2 ( t )

图8. 执行器故障 f a 2 ( t ) 及它的估计 f ^ a 2 ( t )

5. 结论

本文针对具有时变时滞、执行器故障和外部扰动的T-S模糊系统,提出了一种故障估计策略。模糊学习观测器可以很好地刻画故障估计的大小和形状。在本文给出的模糊学习观测器的基础上,构造了一个新的Lyapunov-Krasovskii函数,并以LIMs的形式给出了保守性更小的时滞依赖的稳定性条件。数值动力学模型的仿真结果证明了我们结果的有效性。

参考文献

[1] Sun, S., Wang, Y., Zhang, H. and Xie, X. (2020) A New Method of Fault Estimation and Tolerant Control for Fuzzy Systems against Time-Varying Delay.Nonlinear Analysis:Hybrid Systems, 38, Article 100942.
https://doi.org/10.1016/j.nahs.2020.100942
[2] Qi, W., Zong, G. and Su, S.F. (2022) Fault Detection for Semi-Markov Switching Systems in the Presence of Positivity Constraints.IEEE Transactions on Cybernetics, 52, 13027-13037.
https://doi.org/10.1109/TCYB.2021.3096948
[3] Mu, Y., Zhang, H., Yan, Y. and Xie, X. (2023) Distributed Observer-Based Robust Fault Estimation Design for Discrete-Time Interconnected Systems with Disturbances.IEEE Transactions on Cybernetics, 53, 1-11.
https://doi.org/10.1109/TCYB.2022.3232531
[4] Zhang, H., Mu, Y., Gao, Z. and Wang, W. (2022) Observer-Based Fault Reconstruction and Fault-Tolerant Control for Nonlinear Systems Subject to Simultaneous Actuator and Sensor Faults.IEEE Transactions on Fuzzy Systems, 30, 2971-2980.
https://doi.org/10.1109/TFUZZ.2021.3098341
[5] Wang, R., Jiang, B. and Liu, J. (2018) Fault Estimation and Accommodation for a Class of Nonlinear System Based on Neural Network Observer.Transactions of Nanjing University of Aeronautics & Astronautics, 35, 104-111.
[6] Han, J., Zhang, H., Wang, Y.,et al. (2016) Robust State/Fault Estimation and Fault Tolerant Control for T-S Fuzzy Systems with Sensor and Actuator Faults.Journal of the Franklin Institute, 353, 615-641.
https://doi.org/10.1016/j.jfranklin.2015.12.009
[7] Liu, M., Cao, X. and Shi, P. (2013) Fuzzy-Model-Based Fault-Tolerant Design for Nonlinear Stochastic Systems Against Simultaneous Sensor and Actuator Faults.IEEE Transactions on Fuzzy Systems, 21, 789-799.
https://doi.org/10.1109/TFUZZ.2012.2224872
[8] Naderi, E. and Khorasani, K. (2017) A Data-Driven Approach to Actuator and Sensor Fault Detection, Isolation and Estimation in Discrete-Time Linear Systems.Automatica, 85, 165-178.
https://doi.org/10.1016/j.automatica.2017.07.040
[9] You, F.Q. and Wang, C. (2022) Robust Fault Estimation Based on Proportional Differential Learning Observer for Linear Continuous-Time Systems with State Time-Varying Delay.International Journal of Control,AutomationandSystems, 20, 58-72.
https://doi.org/10.1007/s12555-020-0750-9
[10] Zhang, X.Y. and Cheng, S.Y. (2019) Robust Fault Estimation Based on Proportional Integral Observer for Takagi-Sugeno Fuzzy Systems. 2019Chinese Automation Congress(CAC), Hangzhou, 22-24 November 2019, 834-838.
https://doi.org/10.1109/CAC48633.2019.8996790
[11] Jia, Q.X., Wu, L. and Li, H.Y. (2022) Robust Actuator Fault Reconstruction for Takagi-Sugeno Fuzzy Systems with Unknown Input via a Synthesized Learning and Sliding-Mode Observer.Asian Journal of Control, 19, 799-809.
https://doi.org/10.1007/s12555-019-0747-4
[12] Kien, C.V., Anh, H.P.H. and Nam, N.T. (2018) Cascade Training Multilayer Fuzzy Model for Nonlinear Uncertain System Identification Optimized by Differential Evolution Algorithm.International Journal of Fuzzy Systems, 20, 1671-1684.
https://doi.org/10.1007/s40815-017-0431-x
[13] Hassanabadi, A., Pourdadashi, F., Shafiee, M. and Puig, V. (2022) Simultaneous Actuator and Sensor Fault Reconstruction of Singular Delayed Linear Parameter Varying Systems in the Presence of Unknown Time Varying Delays and Inexact Parameters.International Journal of Adaptive Control and Signal Processing, 36, 3043-3065.
https://doi.org/10.1002/acs.3498
[14] Eddine, C.D. and Noura, M. (2017) H∞ Fault Estimation for Nonlinear Discrete Time-Delay System with Actuator and Sensor Faults Using Nonlinear t-s Fuzzy Models. 2017 5thInternational Conference on Electrical Engineering-Boumerdes(ICEE-B), Algeria, 29-31 October 2017, 1-6.
https://doi.org/10.1109/ICEE-B.2017.8192072
[15] Makni, S., Bouattour, M.,et al. (2017) Robust Adaptive Observer Design for Fast Fault Estimation for Nonlinear TS Fuzzy Systems Using Descriptor Approach. 2017 6thInternational Conference on Systems and Control(ICSC), Batna, 7-9 May 2017, 249-254.
https://doi.org/10.1109/ICoSC.2017.7958708
[16] Huang, S.J. and Yang, G.H. (2016) Fault Estimation for Fuzzy Delay Systems: A Minimum Norm Least Squares Solution Approach.IEEE Transactions on Cybernetics, 47, 2389-2399.
https://doi.org/10.1109/TCYB.2016.2586968
[17] Ziyabari, S.H.S. and Shoorehdeli, M.A. (2017) Robust Fuzzy Fault Estimation Based on Decoupled Transform and Unknown Input Sliding Mode Observer. 2017Iranian Conference on Electrical Engineering(ICEE), Tehran, 2-4 May 2017, 772-777.
[18] Liu, S., Li, X., Wang, H.,et al. (2018) Adaptive Fault Estimation for T-S Fuzzy Systems with Unmeasurable Premise Variables.Advances in Difference Equations, 2018, Article No. 105.
https://doi.org/10.1186/s13662-018-1571-5
[19] Xie, X.P., Yue, D. and Park, J.H. (2018) Robust Fault Estimation Design for Discrete-Time Nonlinear Systems via a Modified Fuzzy Fault Estimation Observer.ISA Transactions, 73, 22-30.
https://doi.org/10.1016/j.isatra.2017.12.007
[20] Wang, M., Qiu, J. and Feng, G. (2019) Event-Triggered State Estimation for T-S Fuzzy Affine Systems Based on Piecewise Lyapunov-Krasovskii Functionals.Control Theory and Technology, 17, 99-111.
https://doi.org/10.1007/s11768-019-8189-3
[21] Zhang, X. and Chen, Y.Q. (2017) A Solid Criterion Based on Strict LMI Without Invoking Equality Constraint for Stabilization of Continuous Singular Systems.ISA Transactions, 71, 272-279.
https://doi.org/10.1016/j.isatra.2017.08.022
[22] Sun, C., Wang, F. and He, X. (2015) Robust Fault Estimation for Takagi-Sugeno Nonlinear Systems with Time-Varying State Delay.Circuits,SystemsandSignal Processing, 34, 641-661.
https://doi.org/10.1007/s00034-014-9855-9
[23] Li, H., You, F., Wang, F.,et al. (2017) Robust Fast Adaptive Fault Estimation and Tolerant Control for T-S Fuzzy Systems with Interval Time-Varying Delay.International Journal of Systems Science, 48, 1708-1730.
https://doi.org/10.1080/00207721.2017.1282057
[24] Bai, L., Tian, Z. and Shi, S. (2006) Design of H∞ Robust Fault Detection Filter for Linear Uncertain Time-Delay Systems.ISA Transactions, 45, 491-502.
https://doi.org/10.1016/S0019-0578(07)60227-4
[25] Jia, Q. and Chen, W. (2016) Fault Reconstruction for Continuous-Time Systems via Learning Observers.Asian Journal of Control, 18, 549-561.
https://doi.org/10.1002/asjc.1076

为你推荐

Baidu
map