文献类型：期刊文献

中文题名：基于Dueling DDQN的无人车换道决策模型

英文题名：Autonomous vehicle lane change strategy model based on dueling double deep Q network

作者：张鑫辰[1];张军[2];刘元盛[2];谢龙洋[1]

第一作者：张鑫辰

机构：[1]北京联合大学北京市信息服务工程重点实验室,北京100101;[2]北京联合大学机器人学院,北京100101

第一机构：北京联合大学北京市信息服务工程重点实验室

年份：2022

卷号：54

期号：1

起止页码：63-71

中文期刊名：东北师大学报：自然科学版

外文期刊名：Journal of Northeast Normal University(Natural Science Edition)

收录：CSTPCD;;北大核心:【北大核心2020】；

基金：国家自然科学基金资助项目(61871039,61871038);北京联合大学人才强校优选计划项目(BPHR2020BZ01);北京联合大学研究生资助项目。

语种：中文

中文关键词：无人车;换道决策;双深度Q网络

外文关键词：autonomous vehicle;lane change strategy;DDQN

摘要：针对高速公路场景中无人驾驶车的换道决策问题,提出一种基于竞争结构的双深度Q网络(DDQN)的无人车换道决策模型.在深度Q网络的基础上,将无人车动作的选择和评估分别用不同的神经网络来实现,并将Q网络分为仅与状态S相关的价值函数和同时与状态S和动作A相关的优势函数两部分,使得Dueling DDQN模型可以更好地理解外部的状态环境.使用训练模型在不同复杂程度的道路环境中进行测试,同时与DQN和DDQN进行了实验对比.结果表明,该算法提高了无人车换道决策的成功率,并在保证车辆安全的前提下提高了无人车的行驶效率,在复杂的道路场景下的适用性更强.
Due to the problem of lane change strategy for autonomous vehicles in the highway, the paper proposed a model for lane change strategy based on a Dueling Double Deep Q Network.Based on the deep Q network, the selection and evaluation of actions are implemented by different neural networks.At the same time, the Q network is divided into a value function which is only related to state S and an advantage function which is related to state S and action A,so that the Dueling DDQN model has a better understanding of the external state environment.The training model is used to test in the road environments with different degrees of complexity, and the results are compared with DQN and DDQN algorithm.The experiments show that the algorithm not only improves the success rate of autonomous vehicles lane change strategy, but also improves the driving efficiency of autonomous vehicles while ensuring vehicle safety.The proposed model is applicable in complex road scenes.