一个基于妥协解的多目标线性规划分类模型
- 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。
∗
1 2
1 100049
2 100190
(MCLP)
MCLP
MCLP
A compromise-based MCLP classification model
Bo Wang1Yong Shi2
(1Graduate University of Chinese Academy of Sciences,Beijing100049)
(2CAS Research Center On Ficitious Economy and Data Science,Beijing100190)
Abstract Although multiple criteria linear programming can deal with classification problem
successfully,an original MCLP model has some difficulties in choosing parameters.To overcome
the problems,compromise-based MCLP model is proposed to offer a good promotion of the
original one.In the latter model,there are also two deviations for every single point;that is,
interior deviation and exterior deviation.Similar to the original MCLP model,for each point,we
want at least one of the deviations to be zero.In addition to modeling work,this paper also gives
a proof of the existence of the parameter selection condition.
Keywords MCLP Interior deviation Exterior deviation Compromise solution
∗ ( 70921061( ),90718042( )) BHP Billiton Co.,Australia
1 (MCLP)
1.1
Ned Freed Fred Glover (Goal Programming,GP) (Multiple Criteria Linear Programming,MCLP)
MCLP
1.1.
1: MCLP
αi βi
∑
min
αi
i
∑
βi
max
i
s.t.A i X=b+αi−βi,A i∈G1,
A i X=b+αi−βi,A i∈G2,
αi,βi 0,i=1,2,...,l
1.2
(Multi-objective Programming,MOP) MOP
max g(x)=(g1(x),g2(x),...,g p(x))
over x∈X.
Pareto Pareto
g i(x) g
i
,i=1,2,...,p.
min P(α,β)
s.t.g i(x)−g
i
=αi−βi,i=1,2,...,p,
αi,βi 0,i=1,2,...,p,
x∈X.
Sawaragi
1.1. α,β R p P(α,β) α β αi βi α∗ β∗
α∗i β∗
i
=0,i=1,2,...,p.
1.
min
p
∑
i=1
a iαi−
p
∑
i=1
b iβi
s.t.g i(x)−g
i
=αi−βi,i=1,2,...,p,
αi,βi 0,i=1,2,...,p,
x∈X.
a i=
b i
min
p
∑
i=1
a i(g i(x)−g
i
) s.t.x∈X.
2. a i>b i,∀i
p
∑i=1a iαi−
p
∑
i=1
b iβi=
p
∑
i=1
b i(αi−βi)+
p
∑
i=1
(a i−b i)αi.
α β a i>b i,∀i P(α,β) α
α∗i β∗
i
=0,i=1,2,...,p.
α∗iβ∗i=0,i=1,2,...,p
1.3
1
n X∗ A i n A i X∗ X∗ l b αi βi α=(α1,α2,...,αl) β=(β1,β2,...,βl) X∗ {A1,A2,...,A l} l α·β=0.
2
2.1
MCLP Shi Yu(1989)
−∑l
i=1
αi α∗ α∗ 0 β∗ 0
∑l
i=1
βi
α∗+∑l
i=1
αi d−α d+α
d−
α
−d+α=α∗+
l
∑
i=1
αi,d−
α
+d+
α
=|α∗+
l
∑
i=1
αi|.
β∗−∑l
i=1
βi d−β d+β
d−
β
−d+
β
=β∗−
l
∑
i=1
βi,d−
β
+d+
β
=|β∗−
l
∑
i=1
βi|.
α∗,β∗ −∑l
i=1
αi
∑l
i=1
βi
|α∗+∑l
i=1
αi| |β∗−
∑l
i=1
βi|
d−
α
+d+
α
+d−
β
+d+
β
.
MCLP
min d−
α+d+
α
+d−
β
+d+
β
s.t.d−
α−d+α=α∗+
l
∑
i=1
αi,