双线性正切制导率
鲁鹏,北京理工大学宇航学院
之前看迭代制导的文章 [1] 时里面提到了双线性正切制导率,很多关于火箭轨迹优化的文章也提到过如较新的文献 [2] 。本文总结双线性正切制导率 (Bi-linear tangent law) 的推导过程 ,分别使用变分法 (calculus of variations) 和最小值原理的基础知识进行推导,主要参考了文献 [3] 的57到58页和文献 [4] 的59页到61页。
一、使用变分法推导
在假设地球重力场为平行力场,重力加速度、推力大小和比冲都为常值条件下,研究最优俯仰角程序。
俯仰平面中的纵向运动方程
x
˙
=
v
x
y
˙
=
v
y
v
˙
x
=
P
m
cos
φ
v
˙
y
=
P
m
sin
φ
−
g
\dot{x} = v_{x}\\ \dot{y} = v_{y} \\ \dot{v}_{x} = \frac{P}{m}\cos\varphi \\ \dot{v}_{y} = \frac{P}{m}\sin\varphi - g
x˙=vxy˙=vyv˙x=mPcosφv˙y=mPsinφ−g
在终点关机时刻
t
k
t_{k}
tk 时,导弹的速度和位置可以表示为
{
v
x
k
=
v
x
0
+
∫
0
t
k
P
m
(
t
)
cos
φ
(
t
)
d
t
v
y
k
=
v
y
0
+
∫
0
t
k
(
P
m
(
t
)
sin
φ
(
t
)
−
g
)
d
t
x
k
=
x
0
+
v
x
0
t
k
+
∫
0
t
k
∫
0
τ
P
m
(
t
)
cos
φ
(
t
)
d
t
d
τ
=
x
0
+
v
x
0
t
k
+
[
τ
∫
0
τ
P
m
(
t
)
cos
φ
(
t
)
d
t
−
∫
0
τ
t
P
m
(
t
)
cos
φ
(
t
)
d
t
]
τ
=
t
k
=
x
0
+
v
x
0
t
k
+
∫
0
t
k
(
t
k
−
t
)
P
m
(
t
)
cos
φ
(
t
)
d
t
y
k
=
y
0
+
v
y
0
t
k
+
∫
0
t
k
(
t
k
−
t
)
(
P
m
(
t
)
sin
φ
(
t
)
−
g
)
d
t
\left\{\begin{aligned} v_{xk}=&v_{x0}+\int_{0}^{t_{k}} \frac{P}{m(t)} \cos \varphi(t) \mathrm{d}t \\ v_{yk}=&v_{y0}+\int_{0}^{t_{k}}\left(\frac{P}{m(t)} \sin \varphi(t)-\mathrm{g}\right) \mathrm{d}t \\ x_{k} =& x_{0}+v_{x0}t_{k}+\int_{0}^{t_{k}}\int_{0}^{\tau}\frac{P}{m(t)} \cos \varphi(t) \mathrm{d}t\mathrm{d}\tau \\ =& x_{0}+v_{x0}t_{k}+ \left[ \tau\int_{0}^{\tau}\frac{P}{m(t)} \cos \varphi(t) \mathrm{d}t - \int_{0}^{\tau}t\frac{P}{m(t)} \cos \varphi(t)\mathrm{d}t \right]_{\tau = t_{k}} \\ =& x_{0}+v_{x0}t_{k}+\int_{0}^{t_{k}}\left(t_{k}-t\right) \frac{P}{m(t)} \cos \varphi(t) \mathrm{d}t \\ y_{k}=&y_{0}+v_{y0} t_{k}+\int_{0}^{t_{k}}\left(t_{k}-t\right)\left(\frac{P}{m(t)} \sin \varphi(t)-g\right) \mathrm{d}t \end{aligned}\right.
⎩⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎧vxk=vyk=xk===yk=vx0+∫0tkm(t)Pcosφ(t)dtvy0+∫0tk(m(t)Psinφ(t)−g)dtx0+vx0tk+∫0tk∫0τm(t)Pcosφ(t)dtdτx0+vx0tk+[τ∫0τm(t)Pcosφ(t)dt−∫0τtm(t)Pcosφ(t)dt]τ=tkx0+vx0tk+∫0tk(tk−t)m(t)Pcosφ(t)dty0+vy0tk+∫0tk(tk−t)(m(t)Psinφ(t)−g)dt
设
φ
(
t
)
\varphi(t)
φ(t)的变分为
δ
φ
(
t
)
\delta\varphi(t)
δφ(t),则对应导弹关机时刻速度泛函和位置泛函的变分为
{
δ
v
x
k
=
−
∫
0
t
k
P
m
sin
φ
(
t
)
δ
φ
(
t
)
d
t
δ
v
y
k
=
∫
0
t
k
P
m
cos
φ
(
t
)
δ
φ
(
t
)
d
t
δ
x
k
=
−
∫
0
t
k
(
t
k
−
t
)
P
m
sin
φ
(
t
)
δ
φ
(
t
)
d
t
δ
y
k
=
∫
0
t
k
(
t
k
−
t
)
P
m
cos
φ
(
t
)
δ
φ
(
t
)
d
t
\left\{\begin{aligned} \delta v_{xk}=&-\int_{0}^{t_{k}} \frac{P}{m} \sin \varphi(t) \delta \varphi(t) \mathrm{d}t \\ \delta v_{yk}=&\int_{0}^{t_{k}} \frac{P}{m} \cos \varphi(t) \delta \varphi(t) \mathrm{d}t \\ \delta x_{k}=&-\int_{0}^{t_{k}}\left(t_{k}-t\right) \frac{P}{m} \sin \varphi(t) \delta \varphi(t) \mathrm{d}t \\ \delta y_{k}=&\int_{0}^{t_{k}}\left(t_{k}-t\right) \frac{P}{m} \cos \varphi(t) \delta \varphi(t) \mathrm{d}t \end{aligned}\right.
⎩⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎧δvxk=δvyk=δxk=δyk=−∫0tkmPsinφ(t)δφ(t)dt∫0tkmPcosφ(t)δφ(t)dt−∫0tk(tk−t)mPsinφ(t)δφ(t)dt∫0tk(tk−t)mPcosφ(t)δφ(t)dt
则射程
L
(
x
k
,
y
k
,
v
x
k
,
v
y
k
)
L(x_{k},y_{k},v_{xk},v_{yk})
L(xk,yk,vxk,vyk) 的变分为
δ
L
=
∂
L
∂
v
x
k
δ
v
x
k
+
∂
L
∂
v
y
k
δ
v
y
k
+
∂
L
∂
x
k
δ
x
k
+
∂
L
∂
y
k
δ
y
k
=
−
∂
L
∂
v
x
k
∫
0
t
k
P
m
sin
φ
(
t
)
δ
φ
(
t
)
d
t
+
∂
L
∂
v
y
k
∫
0
t
k
P
m
cos
φ
(
t
)
δ
φ
(
t
)
d
t
−
∂
L
∂
x
k
∫
0
t
k
(
t
k
−
t
)
P
m
sin
φ
(
t
)
δ
φ
(
t
)
d
t
+
∂
L
∂
y
k
∫
0
t
k
(
t
k
−
t
)
P
m
cos
φ
(
t
)
δ
φ
(
t
)
d
t
=
∫
0
t
k
[
−
∂
L
∂
v
x
k
P
m
sin
φ
(
t
)
+
∂
L
∂
v
y
k
P
m
cos
φ
(
t
)
−
∂
L
∂
x
k
(
t
k
−
t
)
P
m
sin
φ
(
t
)
+
∂
L
∂
y
k
(
t
k
−
t
)
P
m
cos
φ
(
t
)
]
δ
φ
(
t
)
d
t
\begin{aligned} \delta L=& \frac{\partial L}{\partial v_{xk}}\delta v_{xk} + \frac{\partial L}{\partial v_{yk}}\delta v_{yk} + \frac{\partial L}{\partial x_{k}}\delta x_{k} + \frac{\partial L}{\partial y_{k}}\delta y_{k}\\ =& -\frac{\partial L}{\partial v_{xk}} \int_{0}^{t_{k}}\frac{P}{m} \sin \varphi(t) \delta \varphi(t) \mathrm{d}t + \frac{\partial L}{\partial v_{yk}} \int_{0}^{t_{k}}\frac{P}{m} \cos \varphi(t)\delta \varphi(t) \mathrm{d}t \\ & -\frac{\partial L}{\partial x_{k}} \int_{0}^{t_{k}}\left(t_{k}-t\right)\frac{P}{m} \sin \varphi(t)\delta \varphi(t) \mathrm{d}t + \frac{\partial L}{\partial y_{k}} \int_{0}^{t_{k}} \left(t_{k}-t\right)\frac{P}{m} \cos \varphi(t) \delta \varphi(t) \mathrm{d}t \\ =& \int_{0}^{t_{k}}\left[-\frac{\partial L}{\partial v_{xk}} \frac{P}{m} \sin \varphi(t)+\frac{\partial L}{\partial v_{yk}} \frac{P}{m} \cos \varphi(t) \right.\\ & \left.-\frac{\partial L}{\partial x_{k}} \left(t_{k}-t\right)\frac{P}{m} \sin \varphi(t)+\frac{\partial L}{\partial y_{k}} \left(t_{k}-t\right)\frac{P}{m} \cos \varphi(t)\right] \delta \varphi(t) \mathrm{d}t \end{aligned}
δL===∂vxk∂Lδvxk+∂vyk∂Lδvyk+∂xk∂Lδxk+∂yk∂Lδyk−∂vxk∂L∫0tkmPsinφ(t)δφ(t)dt+∂vyk∂L∫0tkmPcosφ(t)δφ(t)dt−∂xk∂L∫0tk(tk−t)mPsinφ(t)δφ(t)dt+∂yk∂L∫0tk(tk−t)mPcosφ(t)δφ(t)dt∫0tk[−∂vxk∂LmPsinφ(t)+∂vyk∂LmPcosφ(t)−∂xk∂L(tk−t)mPsinφ(t)+∂yk∂L(tk−t)mPcosφ(t)]δφ(t)dt
射程达到极值的必要条件是变分
δ
L
=
0
\delta L=0
δL=0 ,即
−
[
∂
L
∂
x
k
(
t
k
−
t
)
+
∂
L
∂
v
x
k
]
sin
φ
(
t
)
+
[
∂
L
∂
y
k
(
t
k
−
t
)
+
∂
L
∂
v
y
k
]
cos
φ
(
t
)
=
0
-\left[\frac{\partial L}{\partial x_{k}}\left(t_{k}-t\right)+\frac{\partial L}{\partial v_{xk}}\right] \sin \varphi(t) +\left[\frac{\partial L}{\partial y_{k}}\left(t_{k}-t\right)+\frac{\partial L}{\partial v_{yk}}\right] \cos \varphi(\mathrm{t})=0
−[∂xk∂L(tk−t)+∂vxk∂L]sinφ(t)+[∂yk∂L(tk−t)+∂vyk∂L]cosφ(t)=0
故最优俯仰角程序为
φ
(
t
)
=
arctan
∂
L
∂
y
k
(
t
k
−
t
)
+
∂
L
∂
V
y
k
∂
L
∂
x
k
(
t
k
−
t
)
+
∂
L
∂
V
x
k
\varphi(t)=\arctan \frac{\frac{\partial L}{\partial y_{k}}\left(t_{k}-t\right)+\frac{\partial L}{\partial V_{yk}} }{ \frac{\partial L}{\partial x_{k}}\left(t_{k}-t\right)+\frac{\partial L}{\partial V_{xk}}}
φ(t)=arctan∂xk∂L(tk−t)+∂Vxk∂L∂yk∂L(tk−t)+∂Vyk∂L
当
∂
L
∂
v
x
k
≠
0
\frac{\partial L}{\partial v_{xk}} \neq 0
∂vxk∂L=0,上式可写成
φ
(
t
)
=
arctan
A
+
B
t
1
+
C
t
\varphi(t)=\arctan \frac{A+B t}{1+C t}
φ(t)=arctan1+CtA+Bt
A、B、C是待定常数,由于控制角
φ
\varphi
φ的正切是两个线性函数的比值,这种制导率就叫做双线性正切制导率。在工程设计中可以直接选择系数 A、B、C 使目标优化。对上式进行一阶泰勒展开可得到俯仰角程序
φ
p
r
(
t
)
\varphi_{pr}(t)
φpr(t) 按时间的线性关系表达为
φ
p
r
(
t
)
=
A
+
B
t
\varphi_{pr}(t) = A+Bt
φpr(t)=A+Bt 。工程设计中也可以使用这种简化形式,直接选择系数 A、B使目标优化。
二、使用最小值原理推导
向直线路径的最大速度转移问题(maximum velocity transfer to a rectilinear path):质量为 m m m的质点上作用着一个大小为 m a ma ma的推力。这里只考虑平面问题,质点在惯性坐标系中的坐标是 x , y x,y x,y,速度分量表示为 u , v u,v u,v,推力的方向用角度 β ( t ) \beta(t) β(t)表示,推力方向是问题中唯一的控制变量
质点的运动方程
u
˙
=
a
cos
β
v
˙
=
a
sin
β
x
˙
=
u
y
˙
=
v
\dot{u} = a\cos\beta \\ \dot{v} = a\sin\beta \\ \dot{x} = u \\ \dot{y} = v
u˙=acosβv˙=asinβx˙=uy˙=v
其中
a
a
a 是时间的已知函数。该问题的哈密度函数
H
=
λ
u
a
cos
β
+
λ
v
a
sin
β
+
λ
x
u
+
λ
y
v
H = \lambda_{u}a\cos\beta + \lambda_{v}a\sin\beta + \lambda_{x}u + \lambda_{y}v
H=λuacosβ+λvasinβ+λxu+λyv
协态方程 (costate equation, adjoint equation)
λ
u
˙
=
∂
H
∂
u
=
−
λ
x
,
λ
v
˙
=
−
λ
y
,
λ
x
˙
=
0
,
λ
y
˙
=
0
\dot{\lambda_{u}} =\frac{\partial H}{\partial u}= -\lambda_{x},\dot{\lambda_{v}} = -\lambda_{y},\dot{\lambda_{x}} = 0,\dot{\lambda_{y}} = 0
λu˙=∂u∂H=−λx,λv˙=−λy,λx˙=0,λy˙=0
对上面的微分方程积分可得
λ
u
=
−
c
1
t
+
c
3
,
λ
v
=
−
c
2
t
+
c
4
,
λ
x
=
c
1
,
λ
y
=
c
2
\lambda_{u}=-c_{1}t+c_{3}, \lambda_{v}=-c_{2}t+c_{4}, \lambda_{x}=c_{1}, \lambda_{y}=c_{2}
λu=−c1t+c3,λv=−c2t+c4,λx=c1,λy=c2
其中
c
1
,
c
2
,
c
3
,
c
4
c_{1},c_{2},c_{3},c_{4}
c1,c2,c3,c4都是常数。
根据最优性条件可知
∂
H
∂
β
=
−
λ
u
a
sin
β
+
λ
v
a
cos
β
=
0
\frac{\partial H}{\partial \beta} = -\lambda_{u}a\sin\beta + \lambda_{v}a\cos\beta = 0
∂β∂H=−λuasinβ+λvacosβ=0
因此最优控制率如下
tan
β
=
λ
v
λ
u
=
−
c
2
t
+
c
4
−
c
1
t
+
c
3
\tan\beta = \frac{\lambda_{v}}{\lambda_{u}} = \frac{-c_{2}t+c_{4}}{-c_{1}t+c_{3}}
tanβ=λuλv=−c1t+c3−c2t+c4
这个就是双线性正切制导率的公式。
现在考虑另一个例子,把这个质点在给定时间 T T T内转移到平行于 x x x轴并且距离 x x x轴 h h h远的路径上,使末端速度水平速度 u ( T ) u(T) u(T)尽可能大,而不对终端 x x x进行限制。
此时这个问题的边界条件如下
u
(
0
)
=
0
λ
u
(
T
)
=
1
v
(
0
)
=
0
v
(
T
)
=
0
,
λ
v
(
T
)
=
ν
v
x
(
0
)
=
0
λ
x
(
T
)
=
0
y
(
0
)
=
0
y
(
T
)
=
h
,
λ
y
(
T
)
=
ν
y
\begin{aligned} u(0) = 0 & & \lambda_{u}(T)=1 \\ v(0) = 0 & & v(T) = 0,\lambda_{v}(T)=\nu_{v} \\ x(0) = 0 & & \lambda_{x}(T) = 0 \\ y(0) = 0 & & y(T) = h,\lambda_{y}(T) = \nu_{y} \end{aligned}
u(0)=0v(0)=0x(0)=0y(0)=0λu(T)=1v(T)=0,λv(T)=νvλx(T)=0y(T)=h,λy(T)=νy
其中
ν
v
,
ν
y
\nu_{v},\nu_{y}
νv,νy 是待确定的拉格朗日乘子。
使用上一部分双线性正切制导率推导过程中的信息可得
λ
x
=
0
\lambda_{x}=0
λx=0,进而
λ
u
=
1
\lambda_{u}=1
λu=1。此时最优控制率变成了线性正切制导率
tan
β
=
tan
β
0
−
c
t
,
where
tan
β
=
ν
v
+
ν
y
T
,
c
=
ν
y
\tan\beta = \tan\beta_{0} - ct, \text{ where }\tan\beta=\nu_{v}+\nu_{y}T,c=\nu_{y}
tanβ=tanβ0−ct, where tanβ=νv+νyT,c=νy
参考:
[1] Chandler D C , Smith I E . Development of the iterative guidance mode with its application to various vehicles and missions.[J]. Journal of Spacecraft & Rockets, 1967, 4(7):898-903.
[2] Federici L , Zavoli A , Colasurdo G , et al. Integrated Optimization of First-Stage SRM and Ascent Trajectory of Multistage Launch Vehicles[J]. Journal of Spacecraft and Rockets, 2021(4):1-12.
[3] 祝强军.弹道导弹弹道仿真与优化设计[D].陕西:西北工业大学,2007,57-58. DOI:10.7666/d.y1033176.
[4] A.E. Bryson, Jr., and Y.-C. Ho. Applied Optimal Control, 2nd edition (revised). Hemisphere Publishing Corp., Washington D.C., 1975
[5] Perkins F M . DERIVATION OF LINEAR-TANGENT STEERING LAWS. 1966.