Heterogeneous Attentions for Solving Pickup and Delivery Problem via Deep Reinforcement Learning
Posted 好奇小圈
tags:
篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了Heterogeneous Attentions for Solving Pickup and Delivery Problem via Deep Reinforcement Learning相关的知识,希望对你有一定的参考价值。
文章目录
前言
论文阅读笔记:Heterogeneous Attentions for Solving Pickup and Delivery Problem via Deep Reinforcement Learning
一、PDP问题
定义:载具从仓库开始,依次访问所有取货节点和配送节点,只执行一次服务,最后返回仓库,尽可能地节省时间。注意,这里PDP允许连续的拾取或交付,或它们的混合,只要它满足优先约束。
n
n
n:客户请求数量
P
=
x
i
i
=
1
n
P=\\ x_i \\_i=1^n
P=xii=1n:pickup node set
D
=
x
i
i
=
n
+
1
2
n
D=\\ x_i \\_i=n+1^2n
D=xii=n+12n:delivery node set
x
i
x_i
xi:pickup node
x
n
+
1
x_n+1
xn+1:delivery node,与 pickup node 为成对的,具有优先级关系
仓库:node 0
X
=
x
0
∪
P
∪
D
X=\\ x_0 \\ \\cup P \\cup D
X=x0∪P∪D:complete node set
X
′
=
x
2
n
+
1
∪
X
=
x
i
i
=
0
2
n
+
1
X'=\\ x_2n+1 \\ \\cup X =\\ x_i \\_i=0^2n+1
X′=x2n+1∪X=xii=02n+1:其中
x
2
n
+
1
x_2n+1
x2n+1为仓库的copy
c
i
\\ c_i \\
ci:包含
x
i
\\ x_i \\
xi二维坐标
v
v
v:载具,从
x
i
x_i
xi全部运输到从
x
i
+
n
x_i+n
xi+n(有无限的运输能力)
D
i
j
D_ij
Dij:表示节点
x
i
x_i
xi到节点
x
j
x_j
xj的欧氏距离
f
f
f为载具
v
v
v的速度
y
i
j
∈
0
,
1
y_ij \\in \\ 0,1 \\
yij∈0,1:二元变量,指示载具
v
v
v是否直接从节点
x
i
x_i
xi到节点
x
j
x_j
xj
B
i
B_i
Bi:表示节点
x
i
x_i
xi的到达时间
M
M
M是一个足够大的数
为追求速度尽量快、距离尽量短,可以得到如下目标函数:
min
∑
i
∈
X
∑
j
∈
X
D
i
j
f
y
i
j
\\min \\sum_i \\in X \\sum_j \\in X \\fracD_i jf y_i j
mini∈X∑j∈X∑fDijyij
约束条件如下:
∑
j
∈
X
y
i
j
=
1
,
i
∈
X
′
∑
i
∈
X
y
i
j
=
1
,
j
∈
X
′
B
j
≥
B
i
+
D
i
j
f
−
M
(
1
−
y
i
j
)
,
i
∈
X
′
,
j
∈
X
′
B
i
+
n
≥
B
i
+
D
i
,
i
+
n
f
,
i
∈
P
,
i
+
n
∈
D
y
i
j
=
0
,
1
,
i
∈
X
′
,
j
∈
X
′
B
i
≥
0
,
i
∈
X
′
\\beginaligned \\sum_j \\in X y_i j &=1, \\quad i \\in X^\\prime\\\\ \\sum_i \\in X y_i j &=1, \\quad j \\in X^\\prime\\\\ B_j & \\geq B_i+\\fracD_i jf-M\\left(1-y_i j\\right), \\quad i \\in X^\\prime, j \\in X^\\prime\\\\ B_i+n & \\geq B_i+\\fracD_i, i+nf, \\quad i \\in P, i+n \\in D\\\\ y_i j &=\\0,1\\, \\quad i \\in X^\\prime, j \\in X^\\prime \\\\ B_i & \\geq 0, \\quad i \\in X^\\prime \\endaligned
j∈X∑yiji∈X∑yijBjBi+nyijBi=1,i∈X′=1,j∈X′≥Bi+fDij−M(1−yij),i∈X′,j∈X′≥Bi+fDi,i+n,i∈P,i+n∈D=0,1,i∈X′,jUbuntu20.04 Attentions
What is Heterogeneous Computing?
目标检测61Dynamic Head Unifying Object Detection Heads with Attentions