基于DQN的自动化集装箱码头自动引导车多目标调度优化
Multiobjective scheduling optimization of AGVs in DQN algorithm-based automated container terminals
初良勇 1梁冬2
作者信息
- 1. 集美大学航海学院,福建厦门 361021;福建航运研究院,福建厦门 361021;集美大学现代物流研究中心,福建厦门 361021
- 2. 集美大学航海学院,福建厦门 361021;集美大学现代物流研究中心,福建厦门 361021
- 折叠
摘要
以自动引导车利用率最大和能源消耗量最小为 目标,建立AGV调度优化数学模型,设计7种不同调度策略作为可变调度策略空间,提炼自动引导车调度问题的状态特征和奖励函数,提出一种基于深度Q学习网络算法(deep Q-network,DQN)的可变调度策略的调度算法.算例分析结果表明:与GA算法和Q-learning算法相比,采用基于DQN的调度优化方法求得的调度方案可使自动引导车的利用率分别提高14.76％和19.92％;能源消耗方面,采用DQN算法求得的调度方案平均能耗与GA算法和Q-learning算法相比分别降低16.88％和10.77％.通过与固定调度策略相比,平均利用率提升12.39％,平均能耗降低7.58％.可见本文方法的求解质量更高,同时与固定策略相比验证了所提出可变策略的有效性.
Abstract
Taking the maximum utilization rate of AGV and the minimum energy consumption as the objectives,we establish a mathematical model of AGV scheduling optimization,design seven different scheduling strategies as the space of variable scheduling strategies,refine the state characteristics and the reward function of the AGV schedu-ling problem,and propose a scheduling algorithm based on the variable scheduling strategy of Deep Q-Network(DQN).Results show that compared with the GA and Q-learning algorithms,the scheduling scheme derived based on the DQN scheduling optimization method can improve the utilization rate of AGVs by 14.76％and 19.92％.For energy consumption,the average energy consumption of the scheduling scheme based on DQN is reduced by 16.88％and 10.77％,compared with that of the GA and Q-learning algorithms.By comparing with the fixed scheduling strategy,the average utilization is improved by 12.39％,and the average energy consumption is reduced by 7.58％.Moreover,the solution quality of the proposed method is higher,while the effectiveness of the proposed variable strategy is verified by comparing it with the fixed strategy.
关键词
自动化集装箱码头/深度强化学习/自动引导车调度/调度策略Key words
automated container terminal/deep reinforcement learning/AGV scheduling/scheduling strategy引用本文复制引用
基金项目
福建省自然科学基金(2021J01820)
国家重点研发计划(2017YFC0805309)
国家社会科学基金重大项目(23&ZD138)
国家社会科学基金重点项目(22AZD108)
福建省新型智库重大项目(24MZKA20)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据