200731
本篇是应用数学之动态最优化理论的笔记,欢迎各位交流!今天是第六部分:确定性最优控制。
6. 确定性最优控制
- 定义哈密顿方程
H ( t , x , u , λ ) = f ( t , x , u ) + λ g ( t , x , u ) H(t, x, u, \lambda)=f(t, x, u)+\lambda g(t, x, u) H(t,x,u,λ)=f(t,x,u)+λg(t,x,u)
6.1 自由端点问题
- 最优控制问题为
max ∫ t 0 t 1 f ( t , x ( t ) , u ( t ) ) d t s.t. x ˙ ( t ) = g ( t , x ( t ) , u ( t ) ) x ( t 0 ) = x 0 \begin{aligned} \max \quad &\int_{t_{0}}^{t_{1}} f(t, x(t), u(t)) \mathrm{d} t\\ \text { s.t. } \quad &\dot{x}(t)=g(t, x(t), u(t))\\ \quad &x\left(t_{0}\right)=x_{0} \end{aligned} max s.t. ∫t0t1f(t,x(t),u(t))dtx˙(t)=g(t,x(t),u(t))x(t0)=x0
其中 u ( t ) u(t) u(t)为控制变量, x ( t ) x(t) x(t)为状态变量。最优化条件需要满足最优性条件(a)、欧拉方程(b)、可行性条件(c)、横截条件(d)和二阶条件(e),分别为
a . H u ( t , x , u , λ ) = f u ( t , x , u ) + λ g u ( t , x , u ) = 0 b . d λ d t = − H x ( t , x , u , λ ) = − f x ( t , x , u ) − λ g x ( t , x , u ) c . d x d t = H λ ( t , x , u , λ ) = g ( t , x , u ) , x ( t 0 ) = x 0 d . λ ( t 1 ) = 0 e . H u u ( t , x , u , λ ) ⩽ 0 ( 最 大 化 问 题 ) H u u ( t , x , u , λ ) ⩾ 0 ( 最 小 化 问 题 ) \begin{array}{l} a. &H_{u}(t, x, u, \lambda)=f_{u}(t, x, u)+\lambda g_{u}(t, x, u)=0\\ b. &\frac{\mathrm{d} \lambda}{\mathrm{d} t}=-H_{x}(t, x, u, \lambda)=-f_{x}(t, x, u)-\lambda g_{x}(t, x, u)\\ c. &\frac{\mathrm{d} x}{\mathrm{d} t}=H_{\lambda}(t, x, u, \lambda)=g(t, x, u), x\left(t_{0}\right)=x_{0}\\ d. &\lambda\left(t_{1}\right)=0\\ e.&H_{u u}(t, x, u, \lambda) \leqslant 0 \quad (最大化问题)\\ &H_{u u}(t, x, u, \lambda) \geqslant 0 \quad (最小化问题) \end{array} a.b.c.d.e.Hu(t,x,u,λ)=fu(t,x,u)+λgu(t,x,u)=0dtdλ=−Hx(t,x,u,λ)=−fx(t,x,u)−λgx(t,x,u)dtdx=Hλ(t,x,u,λ)=g(t,x,u),x(t0)=x0λ(t1)=0Huu(t,x,u,λ)⩽0(最大化问题)Huu(t,x,u,λ)⩾0(最小化问题)
-
若目标函数变为
∫ t 0 t 1 f ( t , x ( t ) , u ( t ) ) d t + φ ( x ( t 1 ) ) \int_{t_{0}}^{t_{1}} f(t, x(t), u(t)) \mathrm{d} t+\varphi\left(x\left(t_{1}\right)\right) ∫t0t1f(t,x(t),u(t))dt+φ(x(t1)) -
只需将横截条件变为
λ ( t 1 ) = φ ′ ( x 1 ) \lambda\left(t_{1}\right)=\varphi^{\prime}\left(x_{1}\right) λ(t1)=φ′(x1) -
若带贴现
∫ t 0 t 1 e − r t f ( t , x ( t ) , u ( t ) ) d t + φ ( x ( t 1 ) ) \int_{t_{0}}^{t_{1}} e^{-r t} f(t, x(t), u(t)) \mathrm{d} t+\varphi\left(x\left(t_{1}\right)\right) ∫t0t1e−rtf(t,x(t),u(t))dt+φ(x(t1)) -
则哈密顿方程变为
H ( t , x , u , λ ) = e − r t [ f ( t , x , u ) + μ g ( t , x , u ) ] = e − r t H ~ , μ = e r t λ H(t, x, u, \lambda)=e^{-r t}[f(t, x, u)+\mu g(t, x, u)]=e^{-r t} \tilde{H} \quad, \mu=e^{r t} \lambda H(t,x,u,λ)=e−rt[f(t,x,u)+μg(t,x,u)]=e−rtH~,μ=ertλ -
称为现值哈密顿方程
-
最优性条件和欧拉方程分别为
H ~ u = 0 μ ˙ = r μ − H ~ x \begin{array}{l} \tilde{H}_{u}=0 \\ \dot{\mu}=r \mu-\tilde{H}_{x} \end{array}