青少年实践

基于DQN强化学习算法的智能网联车辆跟驰模型研究

作者：余君濠¹，张力楠^2,*

¹上海市徐汇中学高二5班上海

²同济大学智能交通运输系统（ITS）研究中心上海

^*标记署名为通讯作者

提交：2025-10-30修改：2025-11-14录用：2025-12-12出版：2026-02-15

少年创新探索 2025, 2(1), 23-30; https://doi.org/10.58244/jle.253448

课题资助：自筹经费。无利益冲突需要说明。

摘要：随着人工智能技术的快速发展，强化学习在智能交通系统中的运用日益广泛。交通跟驰行为是微观交通仿真的核心内容之一，对交通流的稳定性与安全性具有重要影响。本研究基于仿真平台，采用深度Q网络（DQN）强化学习算法，对车辆跟驰行为进行建模与仿真。通过构建仿真环境、设计奖励函数、训练智能体，实现了车辆在复杂交通场景下的自主跟驰行为。实验结果表明，基于DQN的跟驰模型能够有效提升车辆的运行效率与安全性，与传统跟驰模型（如IDM模型）相比，在平均车速稳定性和急刹车次数等指标上表现更优。本研究为理解智能驾驶行为提供了仿真基础，也为智能交通系统的优化提供了新的思路与方法。
关键词：人工智能；交通跟随；DQN强化学习算法

正文内容

可下载并阅读全文PDF，请按照本文版权许可使用。
Download the full text PDF for viewing and using it according to the license of this paper.

参考文献：

Mnih V, Kavukcuoglu K, Silver D, et al. Human-level control through deep reinforcement learning[J]. Nature, 2015, 518(7540): 529-533.
陆百川, 马庆禄, 杜永强. 基于深度强化学习的智能网联汽车跟驰模型[J]. 同济大学学报(自然科学版), 2021, 49(07): 1020-1028.
郭迟，刘经南，左文炜等. 智能网联汽车测试与评价技术[M]. 北京：科学出版社, 2020.
Treiber M, Hennecke A, Helbing D. Congested traffic states in empirical observations and microscopic simulations[J]. Physical review E, 2000, 62(2): 1805.
黄合来, 王鹏, 许项东. 交通仿真技术与应用[M]. 北京: 人民交通出版社, 2018.
Sutton R S, Barto A G. Reinforcement learning: An introduction[J]. 2018.
TESS NG用户手册与技术文档[J/OL]. http://www.tessng.com, 2023.

Research on the car-following model of intelligent connected vehicles based on DQN reinforcement learning algorithm

With the rapid development of artificial intelligence technology, reinforcement learning is increasingly being applied in intelligent transportation systems. Car-following behavior is one of the core contents of microscopic traffic simulation and has a significant impact on the stability and safety of traffic flow. Based on a simulation platform, this study adopts the Deep Q-Network (DQN) reinforcement learning algorithm to model and simulate vehicle car-following behavior. By constructing a simulation environment, designing a reward function, and training an agent, autonomous car-following behavior of vehicles in complex traffic scenarios is achieved. Experimental results show that the car-following model based on DQN can effectively improve the operational efficiency and safety of vehicles. Compared with traditional car-following models (such as the Intelligent Driver Model (IDM)), it performs better in indicators such as average vehicle speed stability and the number of sudden braking events. This study provides a simulation foundation for understanding intelligent driving behavior and offers new ideas and methods for optimizing intelligent transportation systems.
Keywords: Artificial Intelligence; Traffic following; DQN reinforcement learning algorithm