目录
1 Stochastic Programming Models
1.2 Inventory
1.2.1 The Newsvendor Problem
F ( x , d ) = c x + b [ d − x ] + + h [ x − d ] + = max { ( c − b ) x + b d , ( c + h ) x − h d } , b > c F(x, d)=c x+b[d-x]_{+}+h[x-d]_{+}=\max \{(c-b) x+b d,(c+h) x-h d\}, b>c F(x,d)=cx+b[d−x]++h[x−d]+=max{(c−b)x+bd,(c+h)x−hd},b>c
-
d
d
d 已知
Min x ≥ 0 F ( x , d ) \operatorname{Min}_{x \geq 0} F(x, d) Minx≥0F(x,d)- 写成线性规划:
Min x ≥ 0 , v v s.t. v ≥ ( c − b ) x + b d v ≥ ( c + h ) x − h d \begin{array}{rl} \operatorname{Min}_{x \geq 0, v} & v \\ \text { s.t. } & v \geq(c-b) x+b d \\ & v \geq(c+h) x-h d \end{array} Minx≥0,v s.t. vv≥(c−b)x+bdv≥(c+h)x−hd - x ˉ = d \bar{x}=d xˉ=d
- 写成线性规划:
-
d
d
d 未知 --> 随机变量
D
D
D,假设cdf
H
(
⋅
)
H(\cdot)
H(⋅)已知
Min x ≥ 0 { f ( x ) : = E [ F ( x , D ) ] } \operatorname{Min}_{x \geq 0}\{f(x):=\mathbb{E}[F(x, D)]\} Minx≥0{f(x):=E[F(x,D)]}- on average
- E [ F ( x , D ) ] } \mathbb{E}[F(x, D)]\} E[F(x,D)]}是凸的
- x ˉ = H − 1 ( κ ) \bar{x}=H^{-1}(\kappa) xˉ=H−1(κ), κ = b − c b + h \kappa=\frac{b-c}{b+h} κ=b+hb−c (critical ratio)
-
d
d
d 未知,假设上下界已知
Min x ≥ 0 max d ∈ [ l , u ] F ( x , d ) \operatorname{Min}_{x \geq 0} \max _{d \in[l, u]} F(x, d) Minx≥0d∈[l,u]maxF(x,d)- worst case approach
- x ∗ = h l + b u h + b x^{*}=\frac{h l+b u}{h+b} x∗=h+bhl+bu
1.2.2 Chance Constraints
Min x ≥ 0 { f ( x ) : = E [ F ( x , D ) ] } \operatorname{Min}_{x \geq 0}\{f(x):=\mathbb{E}[F(x, D)]\} Minx≥0{f(x):=E[F(x,D)]}
- try to control the risk of the cost F ( x ; D ) F(x;D) F(x;D) to be not “too high”
-
F
(
x
,
D
)
≤
τ
,
∀
d
∈
D
F(x, D) \leq \tau, \forall d \in \mathfrak{D}
F(x,D)≤τ,∀d∈D
- This could be quite restrictive if the uncertainty set D \mathfrak{D} D is large
-
Pr
{
F
(
x
,
D
)
≤
τ
}
≥
1
−
α
\operatorname{Pr}\{F(x, D)\le\tau\}\ge1-\alpha
Pr{F(x,D)≤τ}≥1−α
- a chance constraint