一种基于多智能强化学习的车货匹配算法
A Vehicle-cargo Matching Algorithm Based on Multi-agent Reinforcement Learning
郭振华 1郭钊侠 1王伟2
作者信息
- 1. 四川大学商学院 成都 610064;南通河海大学海洋与近海工程研究院 南通 226334
- 2. 南通河海大学海洋与近海工程研究院 南通 226334;河海大学港口海岸与近海工程学院 南京 210098
- 折叠
摘要
文中针对网络货运企业"多对多"车货匹配问题,基于匹配决策过程,将车货匹配问题建模为一个多智能体马尔科夫决策过程,并利用一种基于多层感知器的代理网络、全连接层的混合网络和超参数网络的多智能体强化学习算法,来模拟并优化多智能体马尔科夫决策过程.基于真实网络货运平台车货匹配数据和不同规模的运输路网环境进行数值实验.结果表明:所提出的算法在解决大规模车货匹配问题具有较好性能.
Abstract
Aiming at the"many-to-many"vehicle-cargo matching problem of network freight enterpri-ses,based on the matching decision-making process,the vehicle-cargo matching problem was modeled as a multi-agent Markov decision-making process.A multi-agent reinforcement learning algorithm based on multi-layer perceptron agent network,fully connected hybrid network and hyperparametric network was used to simulate and optimize the multi-agent Markov decision-making process.Numeri-cal experiments were carried out based on the matching data of real network freight platform and dif-ferent scale transportation network environments.The results show that the proposed algorithm has good performance in solving the large-scale vehicle-cargo matching problem.
关键词
车货匹配/多智能体/马尔科夫决策/强化学习Key words
vehicle-cargo matching/multi-agent/Markov decision/reinforcement learning引用本文复制引用
基金项目
南通市科技项目(社会民生重点)(MS22021002)
国家自然科学基金面上项目(72171159)
出版年
2024
NETL
NSTL
万方数据