强化学习
计算机科学
马尔可夫决策过程
调度(生产过程)
数学优化
作业车间调度
增强学习
马尔可夫过程
人工智能
地铁列车时刻表
数学
统计
操作系统
作者
Min Zhang,Chaohong Pan
出处
期刊:IEEE Transactions on Intelligent Transportation Systems
[Institute of Electrical and Electronics Engineers]
日期:2023-12-21
卷期号:25 (3): 3108-3117
被引量:3
标识
DOI:10.1109/tits.2023.3337334
摘要
How to effectively improve the cargo assembly and multi-vehicle stratified planning has become an urgent problem to be solved. In this paper, Multi-Agent Reinforcement Learning Hierarchical Optimal Scheduling Algorithm (MARLHOSA) is proposed to solve the hierarchical scheduling problem of logistics transport vehicles. We model the hierarchical scheduling problem of logistics transport vehicles as an infinite Markov decision process and set constraints to simulate the actual operating environment. To solve the Markov decision process corresponding to the economic scheduling problem of logistics transport vehicles, this paper uses the close-range strategy optimization algorithm, and uses multi-agent reinforcement learning algorithm based on the clipping mechanism to improve the loss function of the short-range strategy optimization algorithm. In addition, a distributed training architecture was designed for the training process of the close-range strategy optimization algorithm, so as to improve the speed of data collection and training speed and quality. According to a demand order put forward by the company, a path-loading collaborative optimization model was established. After solving the model, the number of vehicles dispatched by each vehicle type according to the optimal path-loading scheme of each vehicle type was determined. The simulation results show that the proposed improved distributed proximity strategy optimization algorithm can achieve the same economic performance as the numerical optimization method. Compared with the traditional algorithm, MARLHOSA can reduce the total vehicle mileage by 34.5% and increase the average loading rate of the carriage by 28.6%. The optimization effect is significant.
科研通智能强力驱动
Strongly Powered by AbleSci AI