eletu, A. (2012). Chance constrained optimization-applications, properties and numerical issues. Retrieved May 24, 2015, from https://www.tu-ilmenau.de/fileadmin/media/simulation/Lehre/Vorlesungsskripte/Lecture_materials_Abebe/CCOPT_talk1.pdf
Henrion, R. (2004). Introduction to Chance-Constrained Programming. Retrieved June 7, 2015, from http://stoprog.org/SPIntro/intro2ccp.php
Henrion, R. (2009). Chance Constrained Problems. Retrieved 7 June 2015, from http://stoprog.org/files/sp_XII_tutorials/Henrion.pdf
Distributionally Robust Chance Constraints
假设
a
~
,
b
~
\tilde{\boldsymbol{a}},\tilde{b}
a~,b~ 是随机变量,
P
\mathbb{P}
P 是它们的一个联合分布。考虑下列形式的机会约束:
P
{
a
~
T
x
+
b
~
≤
0
}
≥
1
−
ϵ
,
ϵ
∈
{
0
,
1
}
.
\mathbb{P} \{ \tilde{\boldsymbol{a}}^T\boldsymbol{x} + \tilde{b} \leq 0 \} \ge 1 - \epsilon, \quad \epsilon\in\{0,1\}.
P{a~Tx+b~≤0}≥1−ϵ,ϵ∈{0,1}. 做适当的变换,可以假定
b
~
\tilde{b}
b~ 始终取值为0, 也就是说,我们只需要考虑下面的约束:
P
{
a
~
T
x
≤
0
}
≥
1
−
ϵ
,
ϵ
∈
{
0
,
1
}
.
\mathbb{P} \{ \tilde{\boldsymbol{a}}^T\boldsymbol{x} \leq 0 \} \ge 1 - \epsilon, \quad \epsilon\in\{0,1\}.
P{a~Tx≤0}≥1−ϵ,ϵ∈{0,1}. 假设具体分布是未知,只知道
P
\mathbb{P}
P 属于某个不确定集
P
\mathcal{P}
P,则有对应的分布鲁棒机会约束问题。
定理. [2] 如果
P
\mathcal{P}
P 是所有均值为
μ
\boldsymbol{\mu}
μ,方差都为
Σ
=
σ
σ
T
\boldsymbol{\Sigma} = \boldsymbol{\sigma} \boldsymbol{\sigma}^T
Σ=σσT 的分布集,则分布鲁棒机会约束问题
inf
P
∈
P
P
{
a
~
T
x
≤
0
}
≥
1
−
ϵ
,
ϵ
∈
{
0
,
1
}
.
\mathop{\inf}\limits_{\mathbb{P}\in\mathcal{P}} \mathbb{P} \{ \tilde{\boldsymbol{a}}^T\boldsymbol{x} \leq 0 \} \ge 1 - \epsilon, \quad \epsilon\in\{0,1\}.
P∈PinfP{a~Tx≤0}≥1−ϵ,ϵ∈{0,1}. 等价于下面的凸二阶锥问题(convex second-order cone constraint):
μ
T
x
+
k
ϵ
σ
T
x
≤
0
\boldsymbol{\mu}^T\boldsymbol{x} + k_{\epsilon} \boldsymbol{\sigma}^T\boldsymbol{x} \leq 0
μTx+kϵσTx≤0 其中
k
ϵ
=
(
1
−
ϵ
)
/
ϵ
k_{\epsilon}=\sqrt{(1-\epsilon)/{\epsilon}}
kϵ=(1−ϵ)/ϵ。
文献
【2】 Calafiore, G. C., & Ghaoui, L. E. (2006). On distributionally robust chance-constrained linear programs. Journal of Optimization Theory and Applications, 130(1), 1–22.