1. 问题描述
给出如下几个条件:
- 数学模型
{ x ˙ 1 = a 11 x 1 + a 12 x 2 + ⋯ + a 1 n x n ⋮ x ˙ i = a i 1 x 1 + a i 2 x 2 + ⋯ + a i n x n ⋮ x ˙ n = a n 1 x 1 + a n 2 x 2 + ⋯ + a n n x n + k u \begin{cases} \dot x_1 = a_{11} x_1 + a_{12} x_2 + \cdots + a_{1n} x_n \\ \vdots \\ \dot x_i = a_{i1} x_1 + a_{i2} x_2 + \cdots + a_{in} x_n \\ \vdots \\ \dot x_n = a_{n1} x_1 + a_{n2} x_2 + \cdots + a_{nn} x_n + \bm{ku} \end{cases} ⎩⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎧x˙1=a11x1+a12x2+⋯+a1nxn⋮x˙i=ai1x1+ai2x2+⋯+ainxn⋮x˙n=an1x1+an2x2+⋯+annxn+ku值得注意的是,只有最后一个式子中含有控制量 u u u。 - 边界条件(共
2
n
\bm{2n}
2n个)
x i ( 0 ) = x i 0 x i ( T ) = x i T x_i(0) = x_{i0} \\ x_i (T) = x_{iT} xi(0)=xi0xi(T)=xiT - 性能指标,即快速作用指标
J = ∫ 0 T 1 d t → min J = \int _0 ^T 1 dt \rightarrow \min J=∫0T1dt→min即 t k − t 0 = T → min t_k - t_0 = T \rightarrow \min tk−t0=T→min。
性能指标即为时间跨度。 - 对控制量的限制
∣ u k ( t ) ∣ ≤ U k max \big\lvert u_k (t) \big\rvert \leq U_{k \max} ∣∣uk(t)∣∣≤Ukmax
2. 解题步骤
首先写出哈密尔顿函数:
H
=
∑
i
=
0
n
ψ
i
(
t
)
⋅
f
i
(
t
)
(1)
H = \sum_{i=0} ^n \psi _i (t) \cdot f_i (t) \tag{1}
H=i=0∑nψi(t)⋅fi(t)(1)其中
ψ
0
\psi_0
ψ0为常数,
ψ
i
≠
0
\psi_i \neq 0
ψi=0,且
ψ
0
=
−
1
<
0
\psi _0 = -1 < 0
ψ0=−1<0关于这一点的疑问可以参考博客优化方法理论合集(8)——庞特里亚金最大值原理。
这样,(1)式亦即
H
(
x
⃗
,
ψ
⃗
,
u
⃗
)
=
ψ
0
⋅
1
+
∑
i
=
1
n
ψ
i
(
t
)
⋅
f
i
(
t
,
x
⃗
,
u
⃗
)
=
−
1
+
∑
i
=
1
n
ψ
i
(
t
)
⋅
f
i
(
t
,
x
⃗
,
u
⃗
)
(2)
\begin{aligned} H \left( \vec x, \vec \psi, \vec u \right) &= \psi_0 \cdot 1 + \sum _{i=1} ^ n \psi _i (t) \cdot f_i \left( t, \vec x, \vec u \right) \\ &= -1 + \sum _{i=1} ^ n \psi _i (t) \cdot f_i \left( t, \vec x, \vec u \right) \tag{2} \end{aligned}
H(x,ψ,u)=ψ0⋅1+i=1∑nψi(t)⋅fi(t,x,u)=−1+i=1∑nψi(t)⋅fi(t,x,u)(2)并依然有
∂
H
∂
u
k
=
0
(3)
\frac{\partial H}{\partial u_k} = 0 \tag{3}
∂uk∂H=0(3)及耦合微分方程组
{
x
˙
i
=
∂
H
∂
ψ
i
ψ
˙
i
=
−
∂
H
∂
x
i
(4)
\begin{cases} \dot x_i = \frac{\partial H}{\partial \psi _i} \\ \dot \psi_i = - \frac{\partial H}{\partial x_i} \end{cases} \tag{4}
{x˙i=∂ψi∂Hψ˙i=−∂xi∂H(4)
在
f
i
f_i
fi中,只有最后一个式子含有
u
u
u,那么
∂
H
∂
u
k
=
ψ
n
(
t
)
⋅
k
(5)
\frac{\partial H}{\partial u_k} = \psi _n (t) \cdot k \tag{5}
∂uk∂H=ψn(t)⋅k(5)(4)式不是
u
u
u的函数(因为在求导过程中
k
u
ku
ku求导变成
k
k
k,
u
u
u消失了)。
而根据(3)又有
∂
H
∂
u
k
=
ψ
n
(
t
)
⋅
k
=
0
(6)
\frac{\partial H}{\partial u_k} = \psi _n (t) \cdot k = 0 \tag{6}
∂uk∂H=ψn(t)⋅k=0(6)而
ψ
i
≠
0
\psi_i \neq 0
ψi=0,因此(6)不成立,此时解题陷入了矛盾。
为解决此矛盾,取
u
(
t
)
=
U
k
max
⋅
s
g
n
(
∂
H
∂
u
)
=
U
k
max
⋅
s
g
n
(
k
ψ
n
(
t
)
)
=
{
+
U
k
max
,
k
ψ
n
(
t
)
>
0
−
U
k
max
,
k
ψ
n
(
t
)
<
0
(7)
\begin{aligned} u(t) &= U_{k \max} \cdot sgn \left( \frac{\partial H}{\partial u} \right) \\ &= U_{k \max} \cdot sgn \left( k \psi _n (t) \right) \\ &= \begin{cases} +U_{k \max}, \quad k \psi_n(t) > 0 \\ -U_{k \max}, \quad k \psi_n(t) < 0 \end{cases} \end{aligned} \tag{7}
u(t)=Ukmax⋅sgn(∂u∂H)=Ukmax⋅sgn(kψn(t))={+Ukmax,kψn(t)>0−Ukmax,kψn(t)<0(7)可以看出,
u
(
t
)
u(t)
u(t)取值仅仅取决于
k
ψ
n
(
t
)
k \psi_n(t)
kψn(t)的符号,当
k
ψ
n
(
t
)
k \psi_n(t)
kψn(t)变号时,
u
(
t
)
u(t)
u(t)也随之改变自己的取值。而
k
k
k的值往往是固定的,因此可以理解为,
u
(
t
)
u(t)
u(t)取值仅仅取决于
ψ
n
(
t
)
\psi_n(t)
ψn(t)的符号,当
ψ
n
(
t
)
\psi_n(t)
ψn(t)变号时,
u
(
t
)
u(t)
u(t)也随之改变自己的取值。
如下图,当 ψ n ( t ) \psi_n(t) ψn(t)变号时, u ( t ) u(t) u(t)也跟着改变自己的取值。
3. 菲尔德鲍姆理论/ n n n–间隔理论
这里不加证明地给出菲尔德鲍姆理论:
若特征方程的根为实根,则
u
(
t
)
u(t)
u(t)变号的次数不超过
(
n
−
1
)
(n-1)
(n−1)次,
n
n
n为特征方程的阶数。
若系统中存在复根,则菲尔德鲍姆理论不再成立, u ( t ) u(t) u(t)变号次数取决于初始条件,因为初始条件不同会导致响应曲线上下的平移,进而改变响应曲线与横轴交点的个数。
4. 例题
给出如下例题:
{
x
˙
1
=
x
2
x
˙
2
=
u
∣
u
(
t
)
∣
≤
U
max
=
1
t
0
=
0
,
t
k
未
知
S
t
a
r
t
=
{
x
1
(
0
)
=
2
x
2
(
0
)
=
1
E
n
d
=
{
x
1
(
t
k
)
=
0
x
2
(
t
k
)
=
0
J
=
∫
o
t
k
1
d
t
→
min
\begin{cases} \dot x_1 = x_2 \\ \dot x_2 = u \end{cases} \\ \big\lvert u(t) \big\rvert \leq U_{\max} = 1 \\ t_0 = 0, \quad t_k 未知 \\ Start = \begin{cases} x_1(0) = 2 \\ x_2 (0) = 1 \end{cases} \\ End = \begin{cases} x_1(t_k) = 0 \\ x_2 (t_k) = 0 \end{cases} \\ J = \int _o ^{t_k} 1 dt \rightarrow \min
{x˙1=x2x˙2=u∣∣u(t)∣∣≤Umax=1t0=0,tk未知Start={x1(0)=2x2(0)=1End={x1(tk)=0x2(tk)=0J=∫otk1dt→min
解:
□
\square \quad
□先根据
x
˙
=
A
x
+
B
u
\dot x = Ax + Bu
x˙=Ax+Bu得出
A
=
[
0
1
0
0
]
A = \left[ \begin{matrix} 0 & 1 \\ 0 & 0 \end{matrix} \right]
A=[0010]求特征根
det
(
I
p
−
A
)
=
∣
p
−
1
0
p
∣
=
p
2
=
0
\det \left( Ip - A \right) = \left| \begin{matrix} p & -1 \\ 0 & p \end{matrix} \right| = p^2 = 0
det(Ip−A)=∣∣∣∣p0−1p∣∣∣∣=p2=0所以
p
1
,
2
=
0
p_{1,2} = 0
p1,2=0,均为实根,则
u
(
t
)
u(t)
u(t)变号次数不超过
(
n
−
1
)
=
2
−
1
=
1
(n-1) = 2-1 = 1
(n−1)=2−1=1次。
ψ 2 ( t ) \psi_2(t) ψ2(t)可能有如下2种情况:
写出哈密尔顿函数
H
=
−
1
+
∑
i
=
1
2
ψ
i
(
t
)
⋅
f
i
(
t
)
=
−
1
+
ψ
1
(
t
)
⋅
f
1
(
t
,
x
⃗
,
u
⃗
)
+
ψ
2
(
t
)
⋅
f
2
(
t
,
x
⃗
,
u
⃗
)
=
−
1
+
ψ
1
(
t
)
⋅
x
2
+
ψ
2
(
t
)
⋅
u
\begin{aligned} H &= -1 + \sum _{i=1} ^2 \psi_i (t) \cdot f_i (t) \\ &= -1 + \psi_1 (t) \cdot f_1 \left( t, \vec x, \vec u \right) + \psi_2 (t) \cdot f_2 \left( t, \vec x, \vec u \right) \\ &= -1 + \psi_1 (t) \cdot x_2 + \psi_2 (t) \cdot u \end{aligned}
H=−1+i=1∑2ψi(t)⋅fi(t)=−1+ψ1(t)⋅f1(t,x,u)+ψ2(t)⋅f2(t,x,u)=−1+ψ1(t)⋅x2+ψ2(t)⋅u
继续写出耦合微分方程组
x
˙
i
=
∂
H
∂
ψ
i
⟹
x
˙
1
=
∂
H
∂
ψ
1
=
x
2
,
x
˙
2
=
∂
H
∂
ψ
2
=
u
\dot x_i = \frac{\partial H}{\partial \psi_i} \\ \Longrightarrow \dot x_1 = \frac{\partial H}{\partial \psi_1} = x_2, \qquad \dot x_2 = \frac{\partial H}{\partial \psi_2} = u
x˙i=∂ψi∂H⟹x˙1=∂ψ1∂H=x2,x˙2=∂ψ2∂H=u
ψ
˙
i
=
−
∂
H
∂
x
i
⟹
ψ
˙
1
=
−
∂
H
∂
x
1
=
0
,
ψ
˙
2
=
−
∂
H
∂
x
2
=
−
ψ
1
(
t
)
\dot \psi_i = - \frac{\partial H}{\partial x_i} \\ \Longrightarrow \dot \psi_1 = - \frac{\partial H}{\partial x_1} = 0, \qquad \dot \psi_2 = - \frac{\partial H}{\partial x_2} = - \psi_1(t)
ψ˙i=−∂xi∂H⟹ψ˙1=−∂x1∂H=0,ψ˙2=−∂x2∂H=−ψ1(t)合并有
{
ψ
˙
1
=
0
ψ
˙
2
=
−
ψ
1
⟹
{
ψ
1
=
C
1
ψ
2
=
−
C
1
t
+
C
2
\begin{cases} \dot \psi_1 = 0 \\ \dot \psi_2 = - \psi_1 \end{cases} \Longrightarrow \begin{cases} \psi_1 = C_1 \\ \psi_2 = -C_1 t + C_2 \end{cases}
{ψ˙1=0ψ˙2=−ψ1⟹{ψ1=C1ψ2=−C1t+C2另一方面,根据式(3)
∂
H
∂
u
=
ψ
2
(
t
)
=
0
\frac{\partial H}{\partial u} = \psi_2(t) = 0
∂u∂H=ψ2(t)=0又一次地,与
ψ
i
≠
0
\psi_i \neq 0
ψi=0相矛盾。
因此,取
u
∘
(
t
)
=
{
+
U
max
=
+
1
,
ψ
2
(
t
)
>
0
−
U
max
=
−
1
,
ψ
2
(
t
)
<
0
u^{\circ} (t) = \begin{cases} +U_{\max} = +1, \quad \psi_2(t) > 0 \\ -U_{\max} = -1, \quad \psi_2(t) < 0 \end{cases}
u∘(t)={+Umax=+1,ψ2(t)>0−Umax=−1,ψ2(t)<0
列出如下2种情况
这两种情况即为可能的控制信号取值。