基础部分
定义:
PD 正定: 除了原点为零点其他都大于零
V ( 0 ) = 0 V ( x ) > 0 , ∀ x ∈ D − { 0 } \begin{aligned}&V(0)=0 \\&V(x)>0, \forall x \in D-\{0\}\end{aligned} V(0)=0V(x)>0,∀x∈D−{0}
PSD 半正定: 除了原点为零点以外还有其他零点
V ( 0 ) = 0 V ( x ) ≥ 0 , ∀ x ∈ D − { 0 } \begin{aligned}&V(0)=0 \\&V(x) \ge0, \forall x \in D-\{0\}\end{aligned} V(0)=0V(x)≥0,∀x∈D−{0}
ND 负定: 除了原点为零点其他都小于零
V ( 0 ) = 0 V ( x ) < 0 , ∀ x ∈ D − { 0 } \begin{aligned}&V(0)=0 \\&V(x)<0, \forall x \in D-\{0\}\end{aligned} V(0)=0V(x)<0,∀x∈D−{0}
NSD 半负定: 除了原点为零点以外还有其他零点
V ( 0 ) = 0 V ( x ) ≤ 0 , ∀ x ∈ D − { 0 } \begin{aligned}&V(0)=0 \\&V(x) \le0, \forall x \in D-\{0\}\end{aligned} V(0)=0V(x)≤0,∀x∈D−{0}
李雅普诺夫稳定性内容:
x ˙ = f ( x ) \dot{x}=f(x) x˙=f(x)
x 是平衡点
→ f ( x = 0 ) , x ˙ ∣ x = 0 = 0 \rightarrow f(x=0),\left.\dot{x}\right|_{x=0}=0 →f(x=0),x˙∣x=0=0
平衡点概念:在初始状态 x=0 状态的变化率为0,该点为稳定点
设V
V : P D V ˙ : N S D V:PD\\\dot{V}:NSD V:PDV˙:NSD
⇒ x=0 是一个稳定点
V : P D V ˙ : N D V:PD\\\dot{V}:ND V:PDV˙:ND
⇒ x=0 是一个渐进稳定点
总结:
如果平衡状态 x e x_{e} xe受到扰动后,仍然停留 x e x_{e} xe在附近,我们就称 x e x_{e} xe在李雅普诺夫意义下是稳定的(Lyapunov stable)。
如果平衡状态 x e x_{e} xe受到扰动后,最终都会收敛到 x e x_{e} xe,我们就称 x e x_{e} xe在李雅普诺夫意义下是渐进稳定的(Asymptotically stable)。
如果平衡状态 x e x_{e} xe受到某种扰动后,状态开始偏离 x e x_{e} xe,我们就称 x e x_{e} xe在李雅普诺夫意义下是不稳定的(Unstable)。
根据李雅普诺夫第一法(间接法),矩阵的特征值决定了系统的稳定性:
系数矩阵与系统稳定性的关系_Amanda1m的博客-CSDN博客_矩阵判断系统稳定性
如何理解李雅普诺夫稳定性分析:
例:
x ˙ 1 = a x 1 = f 1 x ˙ 2 = b x 2 + cos x 1 = f 2 \begin{aligned}&\dot{x}_{1}=a x_{1}=f_{1} \\&\dot{x}_{2}=b x_{2}+\cos x_{1}=f_{2}\end{aligned} x˙1=ax1=f1x˙2=bx2+cosx1=f2
设 V = x 1 2 + x 2 2 V=x_{1}^{2}+x_{2}^{2} V=x12+x22
V ˙ = 2 x 1 x ˙ 1 + 2 x 2 x ˙ 2 = 2 x 1 a x 1 + 2 x 2 ( b x 2 + cos x 1 ) = 2 a x 1 2 + 2 b x 2 2 + 2 x 2 cos x 1 = [ 2 x 1 2 x 2 ] [ x ˙ 1 x ˙ 2 ] = [ ∂ V ∂ x 1 ∂ V ∂ x 2 ] [ f 1 f 2 ] = ∇ V ⋅ f ( x ) = L f V ( x ) \begin{aligned}\dot{V} &=2 x_{1} \dot{x}_{1}+2 x_{2} \dot{x}_{2} \\&=2 x_{1} a x_{1}+2 x_{2}\left(b x_{2}+\cos x_{1}\right) \\&=2 a x_{1}^{2}+2 b x_{2}^{2}+2 x_{2} \cos x_{1} \\&=\left[\begin{array}{ll}2 x_{1} & 2 x_{2}\end{array}\right]\left[\begin{array}{l}\dot{x}_{1} \\\dot{x}_{2}\end{array}\right] \\&=\left[\begin{array}{ll}\frac{\partial V}{\partial x 1} & \frac{\partial V}{\partial x_{2}}\end{array}\right]\left[\begin{array}{l}f_{1} \\f_{2}\end{array}\right] \\&=\nabla V \cdot f(x)=L_{f} V(x)\end{aligned} V˙=2x1x˙1+2x2x˙2=2x1ax1+2x2(bx2+cosx1)=2ax12+2bx22+2x2cosx1=[2x12x2][x˙1x˙2]=[∂x1∂V∂x2∂V][f1f2]=∇V⋅f(x)=LfV(x)
最后一步的 f ( x ) = L f V ( x ) f(x)=L_{f} V(x) f(x)=LfV(x)是求函数的李雅普诺夫导数(Lie Derivate),简称李导数。
若求解后的 V ˙ = 0 ≤ 0 \dot{V}=0\le0 V˙=0≤0说明为半负定NSD,系统稳定,说明 x 1 , x 2 x_{1},x_{2} x1,x2有界,不代表 x 1 , x 2 x_{1},x_{2} x1,x2 → 0。
不变性原理
引入新的理论扩大李雅普诺夫稳定性的判定。
- V > 0 V>0 V>0
- V ˙ ≤ 0 \dot V \le 0 V˙≤0
- V ˙ = 0 → x 1 = 0 , x 2 = 0 \dot V=0 \rightarrow x_{1}=0, x_{2}=0 V˙=0→x1=0,x2=0
例:
x ˙ 1 = x 2 x ˙ 2 = − x 1 − x 2 − ( x 1 + x 2 ) 2 x 2 V = x 1 2 + x 2 2 \begin{aligned}\dot{x}_{1} &=x_{2} \\\dot{x}_{2} &=-x_{1}-x_{2}-\left(x_{1}+x_{2}\right)^{2} x_{2} \\V &=x_{1}^{2}+x_{2}^{2}\end{aligned} x˙1x˙2V=x2=−x1−x2−(x1+x2)2x2=x12+x22
V ˙ = ∇ V ⋅ f ( x ) = [ ∂ V ∂ x 1 ∂ V ∂ x 2 ] [ f 1 f 2 ] = [ 2 x 1 2 x 2 ] [ x 2 − x 1 − x 2 − ( x 1 + x 2 ) 2 x 2 ] = 2 x 1 2 x 2 − 2 x 1 x 2 − 2 x 2 2 − 2 x 2 2 ( x 1 + x 2 ) 2 = − 2 x 2 2 ( 1 + ( x 1 + x 2 ) 2 ) \begin{aligned}\dot{V} &=\nabla V \cdot f(x) \\&=\left[\begin{array}{ll}\frac{\partial V}{\partial x_{1}} & \frac{\partial V}{\partial x_{2}}\end{array}\right]\left[\begin{array}{l}f_{1} \\f_{2}\end{array}\right] \\&=\left[\begin{array}{ll}2 x_{1} & 2 x_{2}\end{array}\right]\left[\begin{array}{c}x_{2} \\-x_{1}-x_{2}-\left(x_{1}+x_{2}\right)^{2} x_{2}\end{array}\right] \\&=2 x_{1} 2 x_{2}-2 x_{1} x_{2}-2 x_{2}^{2}-2 x_{2}^{2}\left(x_{1}+x_{2}\right)^{2} \\&=-2 x_{2}^{2}\left(1+\left(x_{1}+x_{2}\right)^{2}\right)\end{aligned} V˙=∇V⋅f(x)=[∂x1∂V∂x2∂V][f1f2]=[2x12x2][x2−x1−x2−(x1+x2)2x2]=2x12x2−2x1x2−2x22−2x22(x1+x2)2=−2x22(1+(x1+x2)2)
x = [ x 1 , 0 ] V > is PSD, V ˙ is NSD V ˙ = 0 ⇒ x 2 = 0 , x ˙ 2 = 0 , x 1 = 0 \begin{aligned}&x=\left[x_{1}, 0\right] \\&V>\text { is PSD, } \dot{V} \text { is NSD } \\&\dot{V}=0 \Rightarrow x_{2}=0, \dot{x}_{2}=0, x_{1}=0\end{aligned} x=[x1,0]V> is PSD, V˙ is NSD V˙=0⇒x2=0,x˙2=0,x1=0
因此系统渐进稳定
- 注:(0, 0)点为系统的平衡点,该点处一阶导数为0
因此当 x 2 = 0 x_{2}=0 x2=0时, x ˙ 2 = 0 \dot x_{2}=0 x˙2=0
非线性系统稳定性设计
基础反馈系统的稳定性
对于一个系统
x ˙ = f ( x , u ) \dot{x}=f(x, u) x˙=f(x,u)
假设 u 是 x 的函数 ϕ ( x ) \phi(x) ϕ(x):state feedback
x ˙ = f ( x , ϕ ( x ) ) \dot{x}=f(x, \phi(x)) x˙=f(x,ϕ(x))
例1:
x ˙ = f ( x , u ) = a x 2 + u \dot{x}=f(x, u)=a x^{2}+u x˙=f(x,u)=ax2+u
f ( 0 , 0 ) = 0 → f(0,0)=0 \rightarrow f(0,0)=0→平衡点在原点
若希望“0”是渐进稳定平衡点
令 u = − a x 2 − x u=-a x^{2}-x u=−ax2−x,u消除了非线性项且提供了稳定性
则上式变为
x ˙ = − x ⇒ x ( t ) = C e − t \begin{aligned}& \dot{x}=-x \\\Rightarrow & x(t)=C e^{-t}\end{aligned} ⇒x˙=−xx(t)=Ce−t
反馈线性化(Feedback Linearization):非线性系统通过输入线性化。
例2:
x ˙ = x 2 − x 3 + u \dot{x}=x^{2}-x^{3}+u x˙=x2−x3+u
用反馈线性化的方法
u 1 = − x 2 + x 3 − x u_{1}=-x^{2}+x^{3}-x u1=−x2+x3−x
代替方法:李雅普诺夫 直接方法(也叫李雅普诺夫第二法)
李雅普诺夫第二法的基本思想
- 求出系统的能量函数(李雅普诺夫函数) V ( x , t ) V(x,t) V(x,t)——标量函数。
- 能量衰减特性用 V ˙ ( x , t ) \dot V(x,t) V˙(x,t)表示。
- 依据系统的运动方程(状态方程)考察能量函数在运动过程中的变化规律。
- 利用 V ( x , t ) V(x,t) V(x,t)和 V ˙ ( x , t ) \dot V(x,t) V˙(x,t)的符号特性,判断平衡状态稳定性。
令 V = 1 2 x 2 V=\frac{1}{2} x^{2} V=21x2
-
问:V为什么这样设定?
需要让x=0是渐进平衡点,则需满足如下条件:
V ( 0 ) = 0 V ( x ) : P D V ˙ ( x ) : N D V(0)=0 \quad V(x):PD \quad \dot V(x):ND V(0)=0V(x):PDV˙(x):ND
V的函数是由以上几种条件的约束下试出来的
V ( 0 ) = 0 V ( x ) : P D V ( x ) > 0 i n x ≠ 0 V ˙ ( x ) : N D V ˙ ( x ) = ∂ V ∂ x ∂ x ∂ t = x x ˙ = x ( x 2 − x 3 + u ) = x 3 − x 4 + x u V(0)=0\\V(x): P D \quad V(x)>0 \quad in x \neq 0\\\dot{V}(x):ND \quad \dot{V}(x)=\frac{\partial V}{\partial x} \frac{\partial x}{\partial t}=x \dot{x}=x\left(x^{2}-x^{3}+u\right)=x^{3}-x^{4}+x u V(0)=0V(x):PDV(x)>0inx=0V˙(x):NDV˙(x)=∂x∂V∂t∂x=xx˙=x(x2−x3+u)=x3−x4+xu
根据 x 3 x^3 x3的函数图像知道 x 3 x^3 x3不是负定的, x 4 x^4 x4是负定的,因此令u包含 − x 2 -x^2 −x2以消去 x 3 x^3 x3
令
u 2 = − x 2 − x ⇒ V ˙ = − x 4 − x 2 u_{2}=-x^{2}-x \Rightarrow \dot{V}=-x^{4}-x^{2} u2=−x2−x⇒V˙=−x4−x2
或者
u 3 = − x 2 ⇒ V ˙ = − x 4 u_{3}=-x^{2} \Rightarrow \dot{V}=-x^{4} u3=−x2⇒V˙=−x4
注:在控制器设计的时候,尽量避免引入x的高次项,这样会在输出引入一个非常大的项,增大反馈对应的输入。
反馈线性化控制
弹簧滑块拉力构成的动态方程:
m x ¨ + α x 3 = F m \ddot{x}+\alpha x^{3}=F mx¨+αx3=F
目标:改变F使得滑块按指定轨迹移动
令:
F = u F=u F=u输入
x 1 = x x_{1}=x x1=x位移
x 2 = x ˙ 1 x_{2}=\dot x_{1} x2=x˙1速度
让 x 1 → x 1 d x_{1} \rightarrow x_{1d} x1→x1d
x 1 d x_{1d} x1d为规定轨迹
x ˙ 1 = x ˙ = x 2 x ˙ 2 = x ¨ = − α m x 1 3 + 1 m u \begin{aligned}\dot{x}_{1} &=\dot{x}=x_{2} \\\dot{x}_{2} &=\ddot{x}=-\frac{\alpha}{m} x_{1}^{3}+\frac{1}{m} u\end{aligned} x˙1x˙2=x˙=x2=x¨=−mαx13+m1u
通过改变输入u来控制 x ˙ 2 → x 2 → x ˙ 1 → x 1 \dot{x}_{2} \rightarrow x_{2} \rightarrow \dot{x}_{1} \rightarrow x_{1} x˙2→x2→x˙1→x1,这种系统称之为Chain of Integrator
引入误差e:
e = x 1 d − x 1 e=x_{1 d}-x_{1} e=x1d−x1
目标:让 e → 0 e \rightarrow 0 e→0
误差随时间的变化:
e ˙ = x ˙ 1 d − x ˙ 1 = x ˙ 1 d − x 2 \dot{e}=\dot{x}_{1 d}-\dot{x}_{1}=\dot{x}_{1 d}-x_{2} e˙=x˙1d−x˙1=x˙1d−x2
找到Lyapunov函数: V ( e ) V(e) V(e)
V ( e ) : P D V ˙ ( e ) : N D ⇒ e → 0 \begin{aligned}&V(e): P D \\&\dot{V}(e): N D \\&\Rightarrow e \rightarrow 0\end{aligned} V(e):PDV˙(e):ND⇒e→0
设
V 1 = 1 2 e 2 V_{1}=\frac{1}{2} e^{2} V1=21e2
V ˙ 1 = ∂ V ∂ e ∂ e ∂ t = e e ˙ = e ( x ˙ 1 d − x 2 ) \dot{V}_{1}=\frac{\partial V}{\partial e} \frac{\partial e}{\partial t}=e \dot{e}=e\left(\dot{x}_{1 d}-x_{2}\right) V˙1=∂e∂V∂t∂e=ee˙=e(x˙1d−x2)
希望 V ˙ 1 \dot V_{1} V˙1为ND,令
x 2 d = x ˙ 1 d + k 1 e x_{2 d}=\dot{x}_{1 d}+k_{1} e x2d=x˙1d+k1e
此时 V ˙ 1 = − k 1 e 2 \dot{V}_{1}=-k_{1}e^{2} V˙1=−k1e2为负定(ND)函数
新目标: x 2 → x 2 d x_{2} \rightarrow x_{2d} x2→x2d(目的是使误差项最小化)
δ = x 2 d − x 2 \delta=x_{2 d}-x_{2} δ=x2d−x2
V ˙ 1 = e ( x ˙ 1 d − ( x 2 d − δ ) ) = − k 1 e 2 + e δ \dot{V}_{1}=e\left(\dot{x}_{1 d}-\left(x_{2 d}-\delta\right)\right)=-k_{1} e^{2}+e \delta V˙1=e(x˙1d−(x2d−δ))=−k1e2+eδ
δ ˙ = x ˙ 2 d − x ˙ 2 = x ¨ = x ¨ 1 d + k 1 e ˙ − ( − α m x 1 3 + 1 m u ) = x ¨ 1 d + k 1 ( x ˙ 1 d − x 2 ) − ( − α m x 1 3 + 1 m u ) \dot{\delta}=\dot{x}_{2 d}-\dot{x}_{2}=\ddot{x}=\ddot{x}_{1 d}+k_{1} \dot{e}-\left(-\frac{\alpha}{m} x_{1}^{3}+\frac{1}{m} u\right) \\ =\ddot{x}_{1 d}+k_{1}\left(\dot{x}_{1 d}-x_{2}\right)-\left(-\frac{\alpha}{m} x_{1}^{3}+\frac{1}{m} u\right) δ˙=x˙2d−x˙2=x¨=x¨1d+k1e˙−(−mαx13+m1u)=x¨1d+k1(x˙1d−x2)−(−mαx13+m1u)
-
注:代入法的思路
消去中间误差以及新引入的量,只保留状态X,以及给定的轨迹 x 1 d x_{1d} x1d,以及最后一个误差
目标:使 δ , e → 0 \delta,e \rightarrow 0 δ,e→0
-
问:为什么要重新找V函数
因为上式中无法保证 V ˙ 1 \dot{V}_{1} V˙1为ND,需要重新设定V函数
设定 V 2 V_{2} V2的方法类似于上步中 V 1 V_{1} V1的方法,利用导数与原变量的关系凑方程式
V ( e , δ ) : P D V ˙ ( e , δ ) : N D \begin{aligned}&V(e, \delta): P D \\&\dot{V}(e, \delta): N D\end{aligned} V(e,δ):PDV˙(e,δ):ND
令 V 2 = V 1 + 1 2 δ 2 V_{2}=V_{1}+\frac{1}{2} \delta^{2} V2=V1+21δ2
因为 V 1 V_{1} V1, 1 2 δ 2 \frac{1}{2}\delta^2 21δ2都是PD,所以 V 2 V_{2} V2也是PD
V ˙ 2 = V ˙ 1 + δ δ ˙ = − k 1 e 2 + δ ( e + δ ˙ ) \begin{aligned}\dot{V}_{2} &=\dot{V}_{1}+\delta \dot{\delta} \\&=-k_{1} e^{2}+\delta(e+\dot{\delta})\end{aligned} V˙2=V˙1+δδ˙=−k1e2+δ(e+δ˙)
前面的为ND,当然也希望后面的为ND,令 e + δ ˙ = − k 2 δ e+\dot{\delta}=-k_{2} \delta e+δ˙=−k2δ
e + x ¨ 1 d + k 1 ( x ˙ 1 d − x 2 ) − ( − α m x 1 3 + 1 m u ) = − k 2 δ u = m e + m x ¨ 1 d + m k 1 ( x ˙ 1 d − x 2 ) + α x 1 3 + m k 2 δ \begin{gathered}e+\ddot{x}_{1 d}+k_{1}\left(\dot{x}_{1 d}-x_{2}\right)-\left(-\frac{\alpha}{m} x_{1}^{3}+\frac{1}{m} u\right)=-k_{2} \delta \\u=m e+m \ddot{x}_{1 d}+m k_{1}\left(\dot{x}_{1 d}-x_{2}\right)+\alpha x_{1}^{3}+m k_{2} \delta\end{gathered} e+x¨1d+k1(x˙1d−x2)−(−mαx13+m1u)=−k2δu=me+mx¨1d+mk1(x˙1d−x2)+αx13+mk2δ
这是最终的表达式
稳定性验证:
δ = x 2 d − x 2 \delta=x_{2 d}-x_{2} δ=x2d−x2代入 e ˙ = x ˙ 1 d − x ˙ 1 = x ˙ 1 d − x 2 \dot{e}=\dot{x}_{1 d}-\dot{x}_{1}=\dot{x}_{1 d}-x_{2} e˙=x˙1d−x˙1=x˙1d−x2可得
e ˙ = − k 1 e + δ \dot{e}=-k_{1} e+\delta e˙=−k1e+δ
u = m e + m x ¨ 1 d + m k 1 ( x ˙ 1 d − x 2 ) + α x 1 3 + m k 2 δ u=m e+m \ddot{x}_{1 d}+m k_{1}\left(\dot{x}_{1 d}-x_{2}\right)+\alpha x_{1}^{3}+m k_{2} \delta u=me+mx¨1d+mk1(x˙1d−x2)+αx13+mk2δ代入 δ ˙ \dot\delta δ˙表达式可得
δ ˙ = x ¨ 1 d + k 1 ( x ˙ 1 d − x 2 ) + α m x 1 3 − e − x ¨ 1 d − k 1 ( x ˙ 1 d − x 2 ) − α m x 1 3 − k 2 δ = − e − k 2 δ \begin{aligned}\dot{\delta}=& \ddot{x}_{1 d}+k_{1}\left(\dot{x}_{1 d}-x_{2}\right)+\frac{\alpha}{m} x_{1}^{3} \\&-e-\ddot{x}_{1 d}-k_{1}\left(\dot{x}_{1 d}-x_{2}\right)-\frac{\alpha}{m} x_{1}^{3}-k_{2} \delta \\=&-e-k_{2} \delta\end{aligned} δ˙==x¨1d+k1(x˙1d−x2)+mαx13−e−x¨1d−k1(x˙1d−x2)−mαx13−k2δ−e−k2δ
将 e ˙ \dot e e˙和 δ ˙ \dot\delta δ˙对应表达式写成矩阵形式
[ e ˙ δ ˙ ] = [ − k 1 1 − 1 − k 2 ] [ e δ ] \left[\begin{array}{l}\dot{e} \\\dot{\delta}\end{array}\right]=\left[\begin{array}{cc}-k_{1} & 1 \\-1 & -k_{2}\end{array}\right]\left[\begin{array}{l}e \\\delta\end{array}\right] [e˙δ˙]=[−k1−11−k2][eδ]
该系统为线性反馈系统(前面假设的时候k大于0)
λ 1 + λ 2 = Λ = − k 1 − k 2 < 0 λ 1 ⋅ λ 2 = k 1 k 2 + 1 > 0 ⇒ λ 1 , λ 2 < 0 \begin{aligned}&\lambda_{1}+\lambda_{2}=\Lambda=-k_{1}-k_{2}<0 \\&\lambda_{1} \cdot \lambda_{2}=k_{1} k_{2}+1>0 \\&\Rightarrow \lambda_{1}, \lambda_{2}<0\end{aligned} λ1+λ2=Λ=−k1−k2<0λ1⋅λ2=k1k2+1>0⇒λ1,λ2<0
-
问:系统渐进稳定的条件是什么?
矩阵特征值均具有负实部(小于0),系统渐进稳定
平衡点为(0, 0)
[ e ˙ δ ˙ ] = 0 ⇒ [ e δ ] = 0 ⇒ \left[\begin{array}{l}\dot{e} \\\dot{\delta}\end{array}\right]=0 \Rightarrow\left[\begin{array}{l}e \\\delta\end{array}\right]=0 \Rightarrow [e˙δ˙]=0⇒[eδ]=0⇒渐进稳定
15.5非线性反步控制设计
x ˙ 1 = x 1 2 + x 2 x ˙ 2 = x 1 + u \begin{aligned}&\dot{x}_{1}=x_{1}^{2}+x_{2} \\&\dot{x}_{2}=x_{1}+u\end{aligned} x˙1=x12+x2x˙2=x1+u
设计控制器 u → x ˙ 2 → x 2 → x ˙ 1 → x 1 u \rightarrow \dot{x}_{2} \rightarrow x_{2} \rightarrow \dot{x}_{1} \rightarrow x_{1} u→x˙2→x2→x˙1→x1,这种系统称之为Chain of Integrator
step1:
设计 x 2 d x_{2d} x2d(中间输入量),使得 x 1 → x 1 d x_{1} \rightarrow x_{1d} x1→x1d,
引入误差 e = x 1 d − x 1 e=x_{1d}-x_{1} e=x1d−x1, e → 0 e \rightarrow 0 e→0
找到一个Lyapunov函数
V ( e ) = 1 2 e 2 V(e)=\frac{1}{2} e^{2} V(e)=21e2:PD
V ˙ ( e ) = e e ˙ \dot V(e)=e\dot e V˙(e)=ee˙:希望是ND
则有 ⇒ t → ∞ , e → 0 \Rightarrow t \rightarrow \infty, e \rightarrow 0 ⇒t→∞,e→0
e e ˙ = e ( x ˙ 1 d − x 1 2 − x 2 ) e \dot{e}=e\left(\dot{x}_{1 d}-x_{1}^{2}-x_{2}\right) ee˙=e(x˙1d−x12−x2)
设 e ˙ = − k 1 e \dot{e}=-k_{1} e e˙=−k1e,则 e e ˙ = − k 1 e 2 e\dot{e}=-k_{1} e^{2} ee˙=−k1e2是ND,
x 2 d = k 1 e + x ˙ 1 d − x 1 2 x_{2 d}=k_{1} e+\dot{x}_{1 d}-x_{1}^{2} x2d=k1e+x˙1d−x12
-
注:
这里因为 x 2 x_{2} x2满足 e ˙ = − k 1 e \dot{e}=-k_{1} e e˙=−k1e,因此 x 2 = x 2 d x_{2}=x_{2d} x2=x2d,此处用 x 2 d x_{2d} x2d表示
step2:
设计 u u u,使得 x 2 → x 2 d x_{2} \rightarrow x_{2d} x2→x2d
引入新的误差 δ = x 2 d − x 2 \delta=x_{2d}-x_{2} δ=x2d−x2, δ → 0 \delta \rightarrow 0 δ→0
找到一个Lyapunov函数
-
注:需要设计一个什么样的函数
该函数需要包含原变量 e e e以及新变量 δ \delta δ
V ( e , δ ) = 1 2 e 2 + 1 2 δ 2 V(e, \delta)=\frac{1}{2} e^{2}+\frac{1}{2} \delta^{2} V(e,δ)=21e2+21δ2:PD
V ˙ ( e , δ ) = e e ˙ + δ δ ˙ \dot V(e, \delta)=e\dot e+\delta\dot\delta V˙(e,δ)=ee˙+δδ˙:希望是ND
则有 ⇒ t → ∞ , e → 0 \Rightarrow t \rightarrow \infty, e \rightarrow 0 ⇒t→∞,e→0
e e ˙ + δ δ ˙ = e ( x ˙ 1 d − x 1 2 − x 2 ) + δ δ ˙ e \dot{e}+\delta \dot{\delta}=e\left(\dot{x}_{1 d}-x_{1}^{2}-x_{2}\right)+\delta \dot{\delta} ee˙+δδ˙=e(x˙1d−x12−x2)+δδ˙
其中 x 2 = x 2 d − δ = k 1 e + x ˙ 1 d − x 1 2 − δ x_{2}=x_{2 d}-\delta=k_{1} e+\dot{x}_{1 d}-x_{1}^{2}-\delta x2=x2d−δ=k1e+x˙1d−x12−δ
e e ˙ + δ δ ˙ = e ( − k 1 e + δ ) + δ δ ˙ = − k 1 e 2 + δ ( e + δ ˙ ) \begin{aligned}e \dot{e}+\delta \dot{\delta} &=e\left(-k_{1} e+\delta\right)+\delta \dot{\delta} \\&=-k_{1} e^{2}+\delta(e+\dot{\delta})\end{aligned} ee˙+δδ˙=e(−k1e+δ)+δδ˙=−k1e2+δ(e+δ˙)
-
问:上式中 e ˙ \dot e e˙为什么不用 − k 1 e -k_{1}e −k1e代换?
因为此处需要 x 2 x_{2} x2参与计算,如果代入 − k 1 e -k_{1}e −k1e,则默认 x 2 = x 2 d x_{2}=x_{2d} x2=x2d,但此处 δ ˙ = x ˙ 2 d − x ˙ 2 \dot\delta=\dot{x}_{2d}-\dot{x}_{2} δ˙=x˙2d−x˙2,两者不相等
前项为ND,当 e + δ ˙ = − k 2 δ e+\dot{\delta}=-k_{2} \delta e+δ˙=−k2δ,后项也为ND
-
注:ND负定满足的条件
除了原点为零点其他都小于零
e + δ ˙ = e + x ˙ 2 d − x ˙ 2 = e + k 1 e ˙ + x ¨ 1 d − 2 x 1 x ˙ 1 − ( x 1 + u ) = e + k 1 e ˙ + x ¨ 1 d − 2 x 1 ( x 1 2 + x 2 ) − ( x 1 + u ) = e + k 1 ( x ˙ 1 d − x 1 2 − x 2 ) + x ¨ 1 d − 2 x 1 ( x 1 2 + x 2 ) − ( x 1 + u ) = − k 2 δ \begin{aligned}e+\dot{\delta} &=e+\dot{x}_{2 d}-\dot{x}_{2} \\&=e+k_{1} \dot{e}+\ddot{x}_{1 d}-2 x_{1} \dot{x}_{1}-\left(x_{1}+u\right) \\&=e+k_{1} \dot{e}+\ddot{x}_{1 d}-2 x_{1}\left(x_{1}^{2}+x_{2}\right)-\left(x_{1}+u\right) \\&=e+k_{1}\left(\dot{x}_{1 d}-x_{1}^{2}-x_{2}\right)+\ddot{x}_{1 d}-2 x_{1}\left(x_{1}^{2}+x_{2}\right)-\left(x_{1}+u\right) \\&=-k_{2} \delta\end{aligned} e+δ˙=e+x˙2d−x˙2=e+k1e˙+x¨1d−2x1x˙1−(x1+u)=e+k1e˙+x¨1d−2x1(x12+x2)−(x1+u)=e+k1(x˙1d−x12−x2)+x¨1d−2x1(x12+x2)−(x1+u)=−k2δ
u = e + k 1 ( x ˙ 1 d − ( x 1 2 + x 2 ) ) + x ¨ 1 d − 2 x 1 ( x 1 2 + x 2 ) − x 1 + k 2 δ u=e+k_{1}\left(\dot{x}_{1 d}-\left(x_{1}^{2}+x_{2}\right)\right)+\ddot{x}_{1 d}-2 x_{1}\left(x_{1}^{2}+x_{2}\right)-x_{1}+k_{2} \delta u=e+k1(x˙1d−(x12+x2))+x¨1d−2x1(x12+x2)−x1+k2δ
总结:
控制 u u u使得 x 1 x_{1} x1追寻 x 1 d x_{1d} x1d分为两部分
- 通过控制 u u u来控制 x 2 x_{2} x2 ( x 2 → x 2 d x_{2} \rightarrow x_{2d} x2→x2d)
- 通过控制 x 2 x_{2} x2来控制 x 1 x_{1} x1 ( x 1 → x 1 d x_{1} \rightarrow x_{1d} x1→x1d)
设计控制器的时候要反向来做,因此叫back-stepping
非线性自适应控制器
假设参数a为常数
x ˙ = a x 2 + u e ˙ = x ˙ d − x ˙ = x ˙ d − a x 2 − u \begin{gathered}\dot{x}=a x^{2}+u \\\dot{e}=\dot{x}_{d}-\dot{x}=\dot{x}_{d}-a x^{2}-u\end{gathered} x˙=ax2+ue˙=x˙d−x˙=x˙d−ax2−u
估计: a ^ \hat a a^
给i及参数: a ~ \tilde a a~
a ~ ˙ = a ˙ − a ^ ˙ = − a ^ ˙ \dot{\tilde{a}}=\dot{a}-\dot{\hat{a}}=-\dot{\hat{a}} a~˙=a˙−a^˙=−a^˙
Define Lyapunov function:
V ( e , a ~ ) = 1 2 e 2 + 1 2 a ~ 2 V(e, \tilde{a})=\frac{1}{2} e^{2}+\frac{1}{2} \tilde{a}^{2} V(e,a~)=21e2+21a~2
V ( e , a ~ ) V(e, \tilde{a}) V(e,a~)is PD.
V ˙ ( e , a ~ ) = e e ˙ + a ~ a ~ ˙ = e ( x ˙ d − a x 2 − u ) − a ~ a ^ ˙ = e ( − ( a − a ^ ) x 2 − k e ) − a ~ a ^ ˙ = − k e 2 − a ~ ( e x 2 + a ^ ˙ ) \begin{aligned}\dot{V}(e, \tilde{a}) &=e \dot{e}+\tilde{a} \dot{\tilde{a}} \\&=e\left(\dot{x}_{d}-a x^{2}-u\right)-\tilde{a} \dot{\hat{a}} \\&=e\left(-(a-\hat{a}) x^{2}-k e\right)-\tilde{a} \dot{\hat{a}} \\&=-k e^{2}-\tilde{a}\left(e x^{2}+\dot{\hat{a}}\right)\end{aligned} V˙(e,a~)=ee˙+a~a~˙=e(x˙d−ax2−u)−a~a^˙=e(−(a−a^)x2−ke)−a~a^˙=−ke2−a~(ex2+a^˙)
其中令 u = x ˙ d − a ^ x 2 + k e u=\dot{x}_{d}-\hat{a} x^{2}+k e u=x˙d−a^x2+ke,
令 e x 2 + a ^ ˙ = 0 e x^{2}+\dot{\hat{a}}=0 ex2+a^˙=0
V ˙ ( e , a ~ ) = − k e 2 V ˙ ( 0 , a ~ ) = 0 \begin{aligned}\dot{V}(e, \tilde{a}) &=-k e^{2} \\\dot{V}(0, \tilde{a}) &=0\end{aligned} V˙(e,a~)V˙(0,a~)=−ke2=0
因此
V
˙
(
e
,
a
~
)
\dot V(e,\tilde a)
V˙(e,a~)为NSD
a ^ = ∫ 0 t e x 2 d t \hat{a}=\int_{0}^{t} e x^{2} d t a^=∫0tex2dt
u
=
x
˙
d
+
x
2
∫
0
t
e
x
2
d
t
+
k
e
u=\dot{x}_{d}+x^{2} \int_{0}^{t} e x^{2} d t+k e
u=x˙d+x2∫0tex2dt+ke
非线性鲁棒控制 (Robust Control)
滑模控制
上一讲,参数未知,只知道它有界
x ˙ = a x 2 + u \dot{x}=a x^{2}+u x˙=ax2+u
x → x d x d − x = e → 0 \begin{aligned}&x \rightarrow x_{d} \\&x_{d}-x=e \rightarrow 0\end{aligned} x→xdxd−x=e→0
一般形式
x ˙ = f ( x ) + u \dot{x}=f(x)+u x˙=f(x)+u
目标: x → x d x \rightarrow x_{d} x→xd
e ˙ = x ˙ d − x ˙ = x ˙ d − f ( x ) − u ∣ f ( x ) ∣ ≤ ρ ( x ) \dot{e}=\dot{x}_{d}-\dot x=\dot{x}_{d}-f(x)-u \quad|f(x)| \leq \rho(x) e˙=x˙d−x˙=x˙d−f(x)−u∣f(x)∣≤ρ(x)
令 u = x ˙ d + k e + ρ ∣ e ∣ e u=\dot{x}_{d}+k e+\rho \frac{|e|}{e} u=x˙d+ke+ρe∣e∣
∣ e ∣ e \frac{|e|}{e} e∣e∣为符号函数(sgn(e)):
∣ e ∣ e = { 1 , e > 0 0 , e = 0 − 1 , e < 0 \frac{|e|}{e}=\left\{\begin{array}{l}1, e>0 \\0, e=0 \\-1, e<0\end{array}\right. e∣e∣=⎩⎨⎧1,e>00,e=0−1,e<0
证明上面的误差项导数 e ˙ \dot e e˙公式,需要从Lyapunov方程的角度去考虑
V ( e ) = 1 2 e 2 V(e)=\frac{1}{2} e^{2} V(e)=21e2
V ˙ ( e ) = e e ˙ = e ( x ˙ d − f ( x ) − u ) = e e ˙ = e ( x ˙ d − f ( x ) − ( x ˙ d + k e + ρ ∣ e ∣ e ) ) = − k e 2 − e f ( x ) − ρ ∣ e ∣ ≤ − k e 2 + ∣ e ∣ ∣ f ( x ) ∣ − ρ ∣ e ∣ ≤ − k e 2 + ∣ e ∣ ρ − ρ ∣ e ∣ = − k e 2 = − 2 k V ( e ) \begin{aligned}\dot{V}(e) &=e \dot{e}=e\left(\dot{x}_{d}-f(x)-u\right) \\&=e \dot{e}=e\left(\dot{x}_{d}-f(x)-\left(\dot{x}_{d}+k e+\rho \frac{|e|}{e}\right)\right) \\&=-k e^{2}-e f(x)-\rho|e| \\& \leq-k e^{2}+|e||f(x)|-\rho|e| \\& \leq-k e^{2}+|e| \rho-\rho|e| \\&=-k e^{2} \\&=-2 k V(e)\end{aligned} V˙(e)=ee˙=e(x˙d−f(x)−u)=ee˙=e(x˙d−f(x)−(x˙d+ke+ρe∣e∣))=−ke2−ef(x)−ρ∣e∣≤−ke2+∣e∣∣f(x)∣−ρ∣e∣≤−ke2+∣e∣ρ−ρ∣e∣=−ke2=−2kV(e)
V ˙ ( e ) + 2 k V ( e ) ≤ 0 \dot{V}(e)+2 k V(e) \leq 0 V˙(e)+2kV(e)≤0
解上述微分方程不等式,引入松弛变量 S ( t ) > 0 S(t)>0 S(t)>0
V ˙ ( e ) + 2 k V ( e ) + S ( t ) = 0 \dot{V}(e)+2 k V(e)+S(t)=0 V˙(e)+2kV(e)+S(t)=0
微分方程
V ˙ ( e ) + 2 k V ( e ) = − S ( t ) \dot{V}(e)+2 k V(e)=-S(t) V˙(e)+2kV(e)=−S(t)
一阶线性微分方程通解
V ( t ) = V ( 0 ) exp ( − 2 k t ) − exp ( − 2 k t ) ∫ 0 t exp ( − 2 k τ ) S ( τ ) d τ V(t)=V(0) \exp (-2 k t)-\exp (-2 k t) \int_{0}^{t} \exp (-2 k \tau) S(\tau) d \tau V(t)=V(0)exp(−2kt)−exp(−2kt)∫0texp(−2kτ)S(τ)dτ
- 注:不要与误差项e弄混
右项大于0
V ( t ) ≤ V ( 0 ) exp ( − 2 k t ) 1 2 e 2 ( t ) ≤ 1 2 e 2 ( 0 ) exp ( − 2 k t ) 1 2 e 2 ( t ) ≤ 1 2 e 2 ( 0 ) exp ( − 2 k t ) ∣ e ( t ) ∣ ≤ ∣ e ( 0 ) exp ( − k t ) ∣ \begin{aligned}V(t) & \leq V(0) \exp (-2 k t) \\\frac{1}{2} e^{2}(t) & \leq \frac{1}{2} e^{2}(0) \exp (-2 k t) \\\sqrt{\frac{1}{2} e^{2}(t)} & \leq \sqrt{\frac{1}{2} e^{2}(0) \exp (-2 k t)} \\|e(t)| & \leq|e(0) \exp (-k t)|\end{aligned} V(t)21e2(t)21e2(t)∣e(t)∣≤V(0)exp(−2kt)≤21e2(0)exp(−2kt)≤21e2(0)exp(−2kt)≤∣e(0)exp(−kt)∣
因此是指数渐进稳定exponentially Stable
e ˙ = x ˙ d − f ( x ) − ( x ˙ d + k e + ρ ∣ e ∣ e ) = − k e − f ( x ) − ρ ∣ e ∣ e \begin{aligned}\dot{e} &=\dot{x}_{d}-f(x)-\left(\dot{x}_{d}+k e+\rho \frac{|e|}{e}\right) \\&=-k e-f(x)-\rho \frac{|e|}{e}\end{aligned} e˙=x˙d−f(x)−(x˙d+ke+ρe∣e∣)=−ke−f(x)−ρe∣e∣
e ˙ = − k e \dot e=-ke e˙=−ke的相平面:
上式中后两项是控制项,当系统偏离
e
˙
=
−
k
e
\dot e=-ke
e˙=−ke时,控制器会想办法滑到线上,这样一个过程称之为滑模控制。(Sliding Mode)
x ˙ = a x 2 + u \dot{x}=a x^{2}+u x˙=ax2+u
∣ a ∣ ≤ ∣ a ˉ ∣ |a| \leq|\bar{a}| ∣a∣≤∣aˉ∣
∣ f ( x ) ∣ = ∣ a ∣ x 2 ≤ ∣ a ˉ ∣ x 2 < ∣ a ˉ ∣ ( x 2 + 0.1 ) u = k e + x ˙ d + ∣ a ˉ ∣ ( x 2 + 0.1 ) ∣ e ∣ e \begin{gathered}|f(x)|=|a| x^{2} \leq|\bar{a}| x^{2}<|\bar{a}|\left(x^{2}+0.1\right) \\u=k e+\dot{x}_{d}+|\bar{a}|\left(x^{2}+0.1\right) \frac{|e|}{e}\end{gathered} ∣f(x)∣=∣a∣x2≤∣aˉ∣x2<∣aˉ∣(x2+0.1)u=ke+x˙d+∣aˉ∣(x2+0.1)e∣e∣
u为控制器
High Gain高增益 and High Frequency Controller高频
一般形式
x ˙ = f ( x ) + u \dot{x}=f(x)+u x˙=f(x)+u
目标: x → x d x \rightarrow x_{d} x→xd
f ( x ) f(x) f(x)为不确定值且有界, ∣ f ( x ) < ρ ( x ) ∣ |f(x)<\rho(x)| ∣f(x)<ρ(x)∣
误差动态响应:
e ˙ = x ˙ d − x ˙ = x ˙ d − f ( x ) − u \dot{e}=\dot{x}_{d}-\dot{x}=\dot{x}_{d}-f(x)-u e˙=x˙d−x˙=x˙d−f(x)−u
令 u = k e + x ˙ d + u a u x u=k e+\dot{x}_{d}+u_{aux} u=ke+x˙d+uaux
回顾:三种类型的鲁棒控制
-
Sliding Model
u a u x 1 = ρ ∣ e ∣ e u_{aux1}=\rho \frac{|e|}{e} uaux1=ρe∣e∣
滑模控制其实是在两个不同的模式不停的切换
符号函数不是+1就是-1,对于执行器带来了很大的挑战,比如自动驾驶,方向盘突然转向。
特性:
t → ∞ , e → 0 t \rightarrow \infty, e \rightarrow 0 t→∞,e→0
-
High Gain
u a u x 2 = 1 ε ρ 2 e u_{aux2}=\frac{1}{\varepsilon} \rho^{2} e uaux2=ε1ρ2e
High Gain:用足够大的输入去抵消不确定性
特性:
∣ e ( t ) ∣ = ε k |e(t)|=\sqrt{\frac{\varepsilon}{k}} ∣e(t)∣=kε
-
High Frequency
u a u x 3 = ρ 2 e ρ ∣ e ∣ + ε u_{aux3}=\frac{\rho^{2} e}{\rho|e|+\varepsilon} uaux3=ρ∣e∣+ερ2e
High Frequency:
ε = 0 : S l i d i n g M o d e l ε ≠ 0 : u aux 3 < ρ \varepsilon=0 :Sliding Model\\\varepsilon \neq 0: u_{\text {aux} 3}<\rho ε=0:SlidingModelε=0:uaux3<ρ
特性:
∣ e ( t ) ∣ = ε k |e(t)|=\sqrt{\frac{\varepsilon}{k}} ∣e(t)∣=kε
对比:High Frequency控制相对于滑模控制来说,两个不同的模式之间的切换更平滑
稳定性证明
证明High Gain系统的稳定性
令 V = 1 2 e 2 V=\frac{1}{2}e^{2} V=21e2
V ˙ = e e ˙ = − e f ( x ) − k e 2 − 1 ε ρ 2 e 2 ≤ ∣ e ∥ f ( x ) ∣ − k e 2 − 1 ε ρ 2 e 2 ≤ ∣ e ∣ ρ − k e 2 − 1 ε ρ 2 ∣ e ∣ 2 = − k e 2 + ρ ∣ e ∣ ( 1 − 1 ε ρ ∣ e ∣ ) \begin{aligned} \dot{V} &=e \dot{e}=-e f(x)-k e^{2}-\frac{1}{\varepsilon} \rho^{2} e^{2} \\ & \leq|e \| f(x)|-k e^{2}-\frac{1}{\varepsilon} \rho^{2} e^{2} \\ & \leq|e| \rho-k e^{2}-\frac{1}{\varepsilon} \rho^{2}|e|^{2} \\ &=-k e^{2}+\rho|e|\left(1-\frac{1}{\varepsilon} \rho|e|\right) \end{aligned} V˙=ee˙=−ef(x)−ke2−ε1ρ2e2≤∣e∥f(x)∣−ke2−ε1ρ2e2≤∣e∣ρ−ke2−ε1ρ2∣e∣2=−ke2+ρ∣e∣(1−ε1ρ∣e∣)
case 1:
ρ ∣ e ∣ > ε ⇒ 1 ε ρ ∣ e ∣ > 1 ⇒ 1 − 1 ε ρ ∣ e ∣ < 0 ⇒ ρ ∣ e ∣ ( 1 − 1 ε ρ ∣ e ∣ ) < 0 ⇒ V ˙ ≤ − k e 2 ⇒ V ˙ ≤ − 2 k V \begin{aligned}&\rho|e|>\varepsilon \\&\Rightarrow \frac{1}{\varepsilon} \rho|e|>1 \\&\Rightarrow 1-\frac{1}{\varepsilon} \rho|e|<0 \\&\Rightarrow \rho|e|\left(1-\frac{1}{\varepsilon} \rho|e|\right)<0 \\&\Rightarrow \dot{V} \leq-k e^{2} \\&\Rightarrow \dot{V} \leq-2 k V \end{aligned} ρ∣e∣>ε⇒ε1ρ∣e∣>1⇒1−ε1ρ∣e∣<0⇒ρ∣e∣(1−ε1ρ∣e∣)<0⇒V˙≤−ke2⇒V˙≤−2kV
case 2:
ρ ∣ e ∣ ≤ ε ⇒ 1 ε ρ ∣ e ∣ ≤ 1 ⇒ 1 ≥ 1 − 1 ε ρ ∣ e ∣ ≥ 0 ⇒ ε ≥ ρ ∣ e ∣ ≥ ( 1 − 1 ε ρ ∣ e ∣ ) ⇒ V ˙ ≤ − k e 2 + ε ⇒ V ˙ ≤ − 2 k V + ε \begin{aligned}&\rho|e| \leq \varepsilon \\&\Rightarrow \frac{1}{\varepsilon} \rho|e| \leq1 \\&\Rightarrow 1 \geq1-\frac{1}{\varepsilon} \rho|e| \geq0 \\&\Rightarrow \varepsilon \geq \rho|e| \geq \left(1-\frac{1}{\varepsilon} \rho|e|\right) \\&\Rightarrow \dot{V} \leq-k e^{2}+\varepsilon \\&\Rightarrow \dot{V} \leq-2 k V+\varepsilon \end{aligned} ρ∣e∣≤ε⇒ε1ρ∣e∣≤1⇒1≥1−ε1ρ∣e∣≥0⇒ε≥ρ∣e∣≥(1−ε1ρ∣e∣)⇒V˙≤−ke2+ε⇒V˙≤−2kV+ε
求解微分方程不等式,引入 S ( t ) > 0 S(t)>0 S(t)>0
V ˙ + 2 k V = ε − S ( t ) \dot{V}+2 k V=\varepsilon-S(t) V˙+2kV=ε−S(t)
一阶常系数非齐次线性微分方程通解
V ( t ) = V ( 0 ) exp ( − 2 k t ) − exp ( − 2 k t ) ∫ 0 t exp ( − 2 k τ ) S ( τ ) d τ + ε exp ( − 2 k t ) ∫ 0 t exp ( 2 k τ ) d τ 1 2 e 2 ( t ) ≤ 1 2 e 2 ( 0 ) exp ( − 2 k t ) − 0 + ε 2 k ( 1 − exp ( − 2 k t ) ) ∣ e ( t ) ∣ ≤ ∣ e ( 0 ) ∣ exp ( − 2 k t ) + ε k ( 1 − exp ( − 2 k t ) ) \begin{aligned}V(t) &=V(0) \exp (-2 k t)-\exp (-2 k t) \int_{0}^{t} \exp (-2 k \tau) S(\tau) d \tau +\varepsilon \exp (-2 k t) \int_{0}^{t} \exp (2 k \tau) d \tau \\\frac{1}{2} e^{2}(t) & \leq \frac{1}{2} e^{2}(0) \exp (-2 k t)-0+\frac{\varepsilon}{2 k}(1-\exp (-2 k t)) \\|e(t)| & \leq \sqrt{|e(0)| \exp (-2 k t)+\frac{\varepsilon}{k}(1-\exp (-2 k t))}\end{aligned} V(t)21e2(t)∣e(t)∣=V(0)exp(−2kt)−exp(−2kt)∫0texp(−2kτ)S(τ)dτ+εexp(−2kt)∫0texp(2kτ)dτ≤21e2(0)exp(−2kt)−0+2kε(1−exp(−2kt))≤∣e(0)∣exp(−2kt)+kε(1−exp(−2kt))
-
注:这里使用的函数求解法
牛顿-莱布尼茨公式: ∫ a b f ( x ) d x = F ( b ) − F ( a ) = F ( x ) ∣ a b \int_{a}^{b} f(x) d x=F(b)-F(a)=\left.F(x)\right|_{a} ^{b} ∫abf(x)dx=F(b)−F(a)=F(x)∣ab(将变量x换成其他函数不改变积分上下限)
换元法: d ( a x ) = a d ( x ) d(ax)=ad(x) d(ax)=ad(x)
当 t → ∞ t \rightarrow \infty t→∞
∣ e ( t ) ∣ = ε k |e(t)|=\sqrt{\frac{\varepsilon}{k}} ∣e(t)∣=kε
Globally Uniformly Ultimately Bounded (GUUB),该结果为始终有界的结果
ε \varepsilon ε很小的时候,稳态误差很小 e → 0 e \rightarrow 0 e→0,但是输入很大 u a u x 2 = 1 ε ρ 2 e u_{aux2}=\frac{1}{\varepsilon} \rho^{2}e uaux2=ε1ρ2e
因此需要自己权衡误差。
证明High Frequency系统的稳定性
u aux 3 = ρ 2 e ρ ∣ e ∣ + ε u_{\text {aux } 3}=\frac{\rho^{2} e}{\rho|e|+\varepsilon} uaux 3=ρ∣e∣+ερ2e
令 V = 1 2 e 2 V=\frac{1}{2} e^{2} V=21e2
V ˙ = e e ˙ ≤ − k e 2 + ρ ∣ e ∣ − e ρ 2 e ρ ∣ e ∣ + ε = − k e 2 + ρ 2 ∣ e ∣ 2 + ρ 2 ∣ e ∣ ε − ρ 2 e 2 ρ ∣ e ∣ + ε = − k e 2 + ε ρ 2 ∣ e ∣ ρ ∣ e ∣ + ε ⇒ V ˙ ≤ − k e 2 + ε \begin{aligned}\dot{V} &=e \dot{e} \leq-k e^{2}+\rho|e|-e \frac{\rho^{2} e}{\rho|e|+\varepsilon} \\&=-k e^{2}+\frac{\rho^{2}|e|^{2}+\rho^{2}|e| \varepsilon-\rho^{2} e^{2}}{\rho|e|+\varepsilon} \\&=-k e^{2}+\varepsilon \frac{\rho^{2}|e|}{\rho|e|+\varepsilon} \\& \Rightarrow \dot{V} \leq-k e^{2}+\varepsilon\end{aligned} V˙=ee˙≤−ke2+ρ∣e∣−eρ∣e∣+ερ2e=−ke2+ρ∣e∣+ερ2∣e∣2+ρ2∣e∣ε−ρ2e2=−ke2+ερ∣e∣+ερ2∣e∣⇒V˙≤−ke2+ε
和 u a u x 2 u_{aux2} uaux2相同,当 t → ∞ t \rightarrow \infty t→∞
∣ e ( t ) ∣ = ε k |e(t)|=\sqrt{\frac{\varepsilon}{k}} ∣e(t)∣=kε
三种鲁棒控制器的比较
详见上章内容——回顾:三种类型的鲁棒控制
x ˙ = f ( x ) + u \dot{x}=f(x)+u x˙=f(x)+u
∣ a ∣ ≤ 1 |a| \leq1 ∣a∣≤1
目标: x → x d x \rightarrow x_{d} x→xd
f ( x ) = a x 2 ≤ ∣ a ∣ ∣ x ∣ 2 ≤ x 2 < x 2 + 0.1 = ρ f(x)=a x^{2} \leq|a||x|^{2} \leq x^{2}<x^{2}+0.1=\rho f(x)=ax2≤∣a∣∣x∣2≤x2<x2+0.1=ρ
ρ \rho ρ为系统的不确定扰动量