-
Notifications
You must be signed in to change notification settings - Fork 12
/
15-轨迹追踪.typ
264 lines (173 loc) · 6.42 KB
/
15-轨迹追踪.typ
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
#import "@local/scibook:0.1.0": *
#show: doc => conf(
title: "轨迹追踪",
author: ("ivaquero"),
header-cap: "现代控制理论",
footer-cap: "github@ivaquero",
outline-on: false,
doc,
)
= 钟摆系统
<钟摆系统>
== 系统构成
对系统
$ dot.double(ϕ) - g / L ϕ + 1 / L dot(δ) = 0 $
令$x_1 = ϕ, x_2 = dot(ϕ), u = 1 / L dot.double(δ)$
可得
$
mat(delim: "[", dot(x)_1; dot(x)_2) = mat(delim: "[", 0, 1; g/L, 0) mat(delim: "[", x_1; x_2) + mat(delim: "[", 0; - 1) u
$
又令
$ u = -mat(delim: "[", k_1, k_2) mat(delim: "[", x_1; x_2) $
其开环平衡点
$
dot(x)_1 = 0 ⇒ x_2 = 0\
dot(x)_2 = 0 ⇒ x_1 = 0
$
故当$t → ∞$,$x_1, x_2 → 0$。
== 误差函数
设目标值$x_(1 d) = 5$,则误差为$e = x_(1 d) - x_1$,于是有
$ dot(e) = dot(x)_(1 d) - dot(x)_1 = -dot(x)_1 = x_2 $
又
$ dot(x)_2 = g / L x_1 - u = g / L (x_(1 d) - e) - u $
即
$
mat(delim: "[", e ̇; dot(x)_2) =
mat(delim: "[", 0, - 1; - g/L, 0)
mat(delim: "[", e; x_2) +
mat(delim: "[", 0; - 1) u +
mat(delim: "[", 0; g/L x_(1 d))
$
其开环平衡点
$
& dot(e) = 0 ⇒ x_(2 f) = 0\
& dot(x)_2 = 0 ⇒ e_f = x_1 d
$
为改变平衡点,令$u = -mat(delim: "[", k_1, k_2) mat(delim: "[", e_(x_2); x_2) + e_f$,代入,得
$
mat(delim: "[", e ̇; dot(x)_2) = mat(delim: "[", 0, - 1; - g/L + k_1, k_2) mat(delim: "[", e; x_2)
$
此时
- $e_f = 0$
- $x_(2 f) = 0$
设计$k_1, k_2$,令$"Re"("eig"(A_("cl"))) < 0$,于是由特征行列式为$0$,得
$
mat(delim: "[", λ, 1; g/L - k_1, λ - k_2)
&= λ^2 - k_2 λ - g / L + k_1\
&= λ^2 + 2 λ + 1
$
得
- $k_1 = 1 + g / L$
- $k_2 = -2$
代入$u$的表达式,得
$ u = -x_d + (1 + g / L) x_1 + 2 x_2 $
= 弹簧振动阻尼系统
== 系统构成
对弹簧振动阻尼系统
#figure(
image("images/model/vibration.drawio.png", width: 40%),
caption: "弹簧振动阻尼系统",
supplement: "图",
)
- 当目标$𝒙_d = vec(delim: "[", 0, 0)$,为经典 LQR 问题
- 当目标$𝒙_d = vec(delim: "[", 1, 0)$或不为$𝟎$时,为轨迹追踪问题
其离散线性系统方程为
$ 𝒙_([k+1]) = 𝑨 𝒙_([k]) + 𝑩 𝒖_([k]) $ <sys-spring>
若追踪目标为
$ 𝒙_(d[k+1]) = 𝑨_D 𝒙_(d[k]) $
设$𝒙_d$为常数向量,则令
$ 𝒆_([k]) = 𝒙_([k]) - 𝒙_(d[k]) = underbrace(mat(delim: "[", 𝑰, -𝑰), "𝑪ₐ") vec(delim: "[", 𝒙_([k]), 𝒙_(d[k])) $
此时,目标值$𝒆_(d[k]) = 0$,问题转化为 LQR 问题。于是有
$
underbrace(vec(delim: "[", 𝒙_([k+1]), 𝒙_(d[k+1])), "𝒙ₐ[k+1]") = underbrace(dmat(delim: "[", 𝑨, 𝑨_0), "𝑨ₐ") underbrace(vec(delim: "[", 𝒙_([k]), 𝒙_(d[k])), "𝒙ₐ[k]") + underbrace(vec(delim: "[", 𝑩, 𝟎), "𝑩ₐ") 𝒖
$
进一步有
$ 𝒆_([k]) = 𝑪_a 𝒙_(a[k]) $
对应的代价函数为
$
J &= 1 / 2 𝒆^(⊤)_([N]) 𝑺 𝒆_([N]) + 1 / 2 ∑_(k=0)^(N-1) (𝒆^(⊤)_([k]) 𝑸 𝒆_([k]) + 𝒖^(⊤)_([k]) 𝑹 𝒖_([k]))\
&= 1 / 2 (𝑪_a 𝒙_(a[k]))^(⊤)_([N]) 𝑺 (𝑪_a 𝒙_(a[k])) + 1 / 2 ∑_(k=0)^(N-1) (
(𝑪_a 𝒙_(a[k]))^(⊤)_([k]) 𝑸 (𝑪_a 𝒙_(a[k]))_([k]) + 𝒖^(⊤)_([k]) 𝑹 𝒖_([k])
)\
&= 1 / 2 𝒙_(a[k])^(⊤) 𝑺_a 𝒙_(a[k]) + 1 / 2 ∑_(k=0)^(N-1) (𝒙_(a[k])^(⊤) 𝑸_a 𝒙_(a[k]) + 𝒖^(⊤)_([k]) 𝑹 𝒖_([k]))
$ <cost-spring>
上式成功将追踪问题转化为了 LQR 问题。但当$𝑹$过大,则系统将躺平,无法进行追踪。
== 稳态非零常数输入
对轨迹追踪问题,有
$ 𝒙_(a[k+1]) = 𝑨_D 𝒙_(d[k]) $ <aug>
考虑特例,如恒温或匀速控制,此时$𝑨_D = 𝑰$,即
$ 𝒙_(d[k]) = 𝒙_d $
且$𝒙_d$处,系统处于稳态,即输入$𝒖$使系统总能得到$𝒙_d$。于是
$ 𝒙_d = 𝑨 𝒙_d + 𝑩 𝒖_d $
可得
$ (𝑰 - 𝑨)𝒙_d = 𝑩 𝒖_d $
定义稳态误差
$ δ 𝒖_([k]) = 𝒖_([k]) - 𝒖_([d]) $
代入@eqt:sys-spring,可得
$
𝒙_([k+1]) &= 𝑨 𝒙_([k]) + 𝑩(δ 𝒖_([k]) + 𝒖_([d])) \
&= 𝑨 𝒙_([k]) + 𝑩 δ 𝒖_([k]) + (𝑰 - 𝑨) 𝒙_d
$
构造增广矩阵
$
𝒙_(a[
k+1
]) = vec(delim: "[", 𝒙_([k+1]), 𝒙_d) = underbrace(mat(delim: "[", 𝑨, 𝑰 - 𝑨; 𝟎, 𝑰), "𝑨ₐ") vec(delim: "[", 𝒙_([k]), 𝒙_d) + underbrace(vec(delim: "[", 𝑩, 𝟎), "𝑩ₐ") δ 𝒖_([
k
])
$
令
$
𝒆_([k]) = 𝒙_([k]) - 𝒙_d = underbrace(mat(delim: "[", 𝑰, -𝑰), "𝑪ₐ") underbrace(vec(delim: "[", 𝒙_([k]), 𝒙_d), "𝒙ₐ[k]")
$
代入@eqt:cost-spring,得
$
J &= 1 / 2 𝒙^(⊤)_(a[N]) 𝑪^(⊤)_a 𝑺 𝑪_a 𝒙_(a[N]) + 1 / 2 ∑_(k=0)^(N-1) (
𝒙^(⊤)_(a[k]) 𝑪^(⊤)_a 𝑸 𝑪_a 𝒙_(a[k]) + δ 𝒖^(⊤)_([k]) 𝑹 δ 𝒖_([k])
)
$ <cost-spring2>
这就得到了一个新的 LQR 问题。
#figure(
image("images/block/lqr-trk-const.drawio.png", width: 40%),
caption: "轨迹追踪 LQR 系统",
supplement: "图",
)
== 稳态非零矩阵输入
对非常数输入,@eqt:aug 中$𝑨_D ≠ 𝑰$。此时,定义输入增量
$ Δ 𝒖_([k]) = underbrace(𝒖_([k]) - 𝒖_([k-1]), "平滑输入的变化") $
代人@eqt:sys-spring,可得
$ 𝒙_([k+1]) = 𝑨 𝒙_([k]) + 𝑩 𝒖_([k]) + 𝑩 𝒖_([k-1]) $
设增广向量
$ 𝒙_(a[k]) = vec(delim: "[", 𝒙_([k]), 𝒙_([d]), 𝒖_([k-1])) $
于是
$ 𝒆_([k]) = 𝒙_([k]) - 𝒙_d = mat(delim: "[", 𝑰, -𝑰, 𝟎) vec(delim: "[", 𝒙_([k]), 𝒙_([d]), 𝒖_([k-1])) = 𝑪_a 𝒙_(a[k]) $
进一步有
$
𝒙_(a[k+1]) = underbrace(mat(delim: "[", 𝑨, 𝟎, 𝑩; 𝟎, 𝑨_D, 𝟎; 𝟎, 𝟎, 𝑰), "𝑨ₐ") 𝒙_(a[
k
]) + underbrace(vec(delim: "[", 𝑩, 𝟎, 𝑰), "𝑩ₐ") Δ 𝒖_([k])
$
类比@eqt:cost-spring2,得
$
J = 1 / 2 𝒙^(⊤)_(a[N]) 𝑪^(⊤)_a 𝑺 𝑪_a 𝒙_(a[N]) + 1 / 2 ∑_(k=0)^(N-1) (
𝒙^(⊤)_(a[k]) 𝑪^(⊤)_a 𝑸 𝑪_a 𝒙_(a[k]) + Δ 𝒖^(⊤)_([k]) 𝑹 Δ 𝒖_([k])
)
$
#figure(
image("images/block/lqr-trk-var.drawio.png", width: 40%),
caption: "稳态非零矩阵输入",
supplement: "图",
)
这里,我们通过矩阵变换将追踪(tracking)转换为了调控(regulation)。
#pagebreak()
== 稳态正弦函数输入
在符合 Newton 第二定律的系统中,正弦振动是线性系统的内生特性
$
x_(1d) &= sin(ω t) \
x_(2d) &= dv(x_(1d), t) = ω cos(ω t) \
dv(x_(2d), t) &= -ω^2 sin(ω t) = -ω^2 x_(1d)
$
其矩阵形式为
$
dv(, t) vec(delim: "[", x_(1d), x_(2d)) = underbrace(mat(delim: "[", 0, 1; -ω^2, 0), 𝐀_D) vec(delim: "[", x_(1d), x_(2d))
$