一种用于天基低轨卫星网络边缘计算的GA-DDPG卸载算法.pdfVIP

工程科学学报，第

46

卷，第

2

期：343−353，2024

年

2

月

Chinese

Journal

of

Engineering,

Vol.

46,

No.

2:

343−353,

February

2024

/10.13374/j.issn2095-9389.2002;

一种用于天基低轨卫星网络边缘计算的GA−DDPG卸

载算法

史栋元，王丽娜✉

北京科技大学计算机与通信工程学院通信工程系，北京

100083

✉通信作者，

E-mail:

wln_ustb@126.com

摘

要

低轨卫星网络是第六代移动通信系统（6G）网络的重要组成部分，弥补了地面基站覆盖的盲区.

由于星上计算能力和

电池容量受限，导致任务出现时延长和能耗高的问题，因此在低轨卫星网络中引入边缘计算，边缘计算的一项关键技术就是

计算卸载.

针对计算卸载过程中星间环境动态变化和高维动作空间的问题，提出一种基于遗传算法（GA）和深度确定性策略

梯度（DDPG）的天基低轨卫星网络边缘计算卸载算法——GA−DDPG算法.

卫星边缘计算环境的不断变化会导致DDPG奖励

稀疏和探索性不足，将GA引入到DDPG算法中，首先，利用GA的选择算子使DDPG算法能够适应不断变化的卫星环境；然

后，针对动作空间维度变大导致DDPG算法收敛不稳定的问题，利用GA种群的多样化探索和种群的冗余提升DDPG算法收

敛的稳定性.

仿真结果表明，GA−DDPG卸载算法能够降低天基低轨卫星网络计算负载，且时延和能耗均低于DDPG卸载算

法和GA卸载算法.

与DDPG卸载算法相比，GA−DDPG卸载算法还能提升收敛速度和稳定性.

关键词

低轨卫星边缘计算；任务卸载；遗传算法；深度强化学习；星间链路

分类号

TN915.9

GA−DDPG

unloading

algorithm

for

edge

computing

in

space-based

LEO

satellite

networks

SHIDongyuan，WANGLina✉

Department

of

Communication

Engineering,

School

of

Computer

and

Communication

Engineering,

University

of

Science

and

Technology

Beijing,

Beijing

100083,

China

✉Corresponding

author,

E-mail:

wln_ustb@126.com

ABSTRACT

Low-earth

orbit

(LEO)

satellite

networks

are

an

important

part

of

the

sixth-generation

mobile

communication

system

(6G)

network,

which

overcomes

the

blind

spots

in

ground-based

station

coverage.

However,

the

limited

onboard

computing

capability

and

battery

capacity

cause

the

problems

of

extended

mission

duration

and

high-energy

consumption;

therefore,

edge

computing

is

introduced

in

LEO

satellite

networks,

and

its

key

technology

is

computational

offloading.

To

address

the

problems

of

dynamic

changes

in

the

intersatellite

environment

and

high-dimensional

action

space

during

computational

offloading,

we

propose

a

genetic

algorithm

(GA)

and

deep

deterministic

policy

gradient

(DDPG)-based

offloading

algorithm

for

edge

computing

in

space-based

LEO

satellite

networks—the

GA-DDPG

algorithm.

The

constant

change

in

a

satellite

edge

computing

environment

will

result

in

sparse