Dynamic On-Demand Crowdshipping Using Constrained and Heuristics-Embedded Double Dueling Deep Q-Network

启发式 计算机科学 强化学习 启发式 数学优化 动作(物理) 运筹学 人工智能 工程类 数学 量子力学 操作系统 物理
作者
Nahid Parvez Farazi,Bo Zou,Theja Tulabandhula
出处
期刊:Transportation Research Part E-logistics and Transportation Review [Elsevier BV]
卷期号:166: 102890-102890 被引量:8
标识
DOI:10.1016/j.tre.2022.102890
摘要

This paper proposes a deep reinforcement learning (DRL)-based approach to the dynamic on-demand crowdshipping problem in which requests constantly arrive in a crowdshipping system for pickup and delivery within limited time windows. The request pickup and delivery are performed by crowdsourcees, who are ordinary people dynamically arriving in and leaving the crowdshipping system, and dedicating their limited and heterogeneous available time and carrying capacity to crowdshipping. In return, crowdsourcees get paid by the delivery service provider who periodically assigns requests to crowdsourcees in the course of a day to minimize shipping cost. We adopt heuristics-embedded Deep Q-Network (DQN) algorithms that incorporate double and dueling structures, to train DRL agents. The idea of heuristics-embedded training is conceived by designing an elaborate action space where several refined local search heuristics are embedded to direct the specific action to take once an action type is chosen by DRL, with the purpose of preserving tractability of DRL training. To tackle the hard constraints pertaining to crowdsourcee and request time windows, we propose and integrate three new strategies (feasibility enforced local search, multiple schedules with different penalties, and exponential penalty) as part of the DRL training and testing. Extensive numerical analysis is conducted and shows that Double Dueling DQN with the exponential penalty strategy demonstrates the best performance. We compare the performance of the agent trained by Double Dueling DQN with conventional heuristic approaches, and find that the agent yields total shipping costs that are on average 24–37% lower than the conventional heuristic approaches. For problem instances that can be solved to optimality, the optimality gap using the trained agent is also quite small, in the range of 3–7%. Moreover, the trained agent is robust to stationary/non-stationary demand patterns. Lastly, our approach is further compared with a recent study that uses heuristics-embedded DQN, and shows superior performance (total shipping costs on average 19% lower) as a result of several differences.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
更新
PDF的下载单位、IP信息已删除 (2025-6-4)

科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
刚刚
2秒前
赘婿应助Husayn采纳,获得10
3秒前
Akim应助wgr采纳,获得10
3秒前
hanch发布了新的文献求助10
4秒前
4秒前
5秒前
7秒前
pengjiejie发布了新的文献求助10
8秒前
MW完成签到,获得积分10
8秒前
xiaomi完成签到,获得积分10
9秒前
支半雪发布了新的文献求助10
9秒前
10秒前
11秒前
11秒前
毕业比耶发布了新的文献求助10
12秒前
着急的cc完成签到,获得积分10
12秒前
13秒前
30888136发布了新的文献求助10
14秒前
15秒前
希望天下0贩的0应助bruce233采纳,获得10
15秒前
www完成签到 ,获得积分10
15秒前
Jasper应助Leon_nomoreLess采纳,获得10
17秒前
Yumori关注了科研通微信公众号
17秒前
Hello应助33采纳,获得10
17秒前
许鹤缤发布了新的文献求助10
17秒前
郭小燕发布了新的文献求助10
18秒前
hanch完成签到,获得积分10
18秒前
量子星尘发布了新的文献求助10
20秒前
21秒前
22秒前
务实发夹完成签到,获得积分10
22秒前
22秒前
周少发布了新的社区帖子
22秒前
智博36完成签到,获得积分10
24秒前
25秒前
醋溜滑板完成签到 ,获得积分10
25秒前
在水一方应助钱钱钱采纳,获得10
25秒前
Yumori发布了新的文献求助10
27秒前
27秒前
高分求助中
A new approach to the extrapolation of accelerated life test data 1000
Picture Books with Same-sex Parented Families: Unintentional Censorship 700
ACSM’s Guidelines for Exercise Testing and Prescription, 12th edition 500
Nucleophilic substitution in azasydnone-modified dinitroanisoles 500
不知道标题是什么 500
Indomethacinのヒトにおける経皮吸収 400
Effective Learning and Mental Wellbeing 400
热门求助领域 (近24小时)
化学 材料科学 医学 生物 工程类 有机化学 生物化学 物理 内科学 纳米技术 计算机科学 化学工程 复合材料 遗传学 基因 物理化学 催化作用 冶金 细胞生物学 免疫学
热门帖子
关注 科研通微信公众号,转发送积分 3975922
求助须知:如何正确求助?哪些是违规求助? 3520226
关于积分的说明 11201711
捐赠科研通 3256720
什么是DOI,文献DOI怎么找? 1798423
邀请新用户注册赠送积分活动 877576
科研通“疑难数据库(出版商)”最低求助积分说明 806452