摘要
针对已有的DQN配时算法无法应用在倒计时交叉口的问题,提出以Webster配时法计算绿信比方案,并将其作为控制动作,以归一化车流量、车速、排队长度构造状态矩阵,通过3层全连接神经网络计算动作价值,引入零奖赏延误因子将延误时间转换为奖励值,使用探索率余弦衰减的方式提高收敛能力,使用Huber函数计算误差提高收敛稳定性,最后利用Vissim仿真平台进行测试。结果表明,该方法在延误指标上相比20种固定配时方案均有不同程度的降低,证明了该方法的有效性。
The existing DQN timing algorithm can not be applied to the problem of countdown intersection. The Webster timing method is proposed to calculate the green signal ratio, and it is used as a control action. Normalized vehicle flow, speed and queue length are used to construct the state matrix. The action value was calculated by 3-layer fully connected neural network. A zero reward delay factor is introduced to convert delay time into reward value. Cosine attenuation of exploration rate is used to improve the convergence ability. Using Huber function to calculate error and improve convergence stability. Finally, the VISSIM simulation platform was used for testing. The results show that the delay index of the proposed method is reduced to different degrees compared with the 20 fixed timing schemes, which proves the effectiveness of the proposed method.
作者
李珊
任安虎
白静静
Li Shan;Ren Anhu;Bai Jingjing(School of Electronic and Information Engineering,Xi’an Technological University,Xi’an 710021,China)
出处
《国外电子测量技术》
北大核心
2021年第10期91-97,共7页
Foreign Electronic Measurement Technology
基金
陕西省科技厅项目(2018GY-153)
陕西省西安市未央区科技局项目(201833)资助。
关键词
智能交通
信号配时
深度强化学习
DQN算法
VISSIM仿真
intelligent transportation
signal timing
deep reinforcement learning
DQN algorithm
vissim simulation
作者简介
李珊,硕士研究生,主要研究方向为智能交通信号控制、交通状态监测等。E-mail:1624693146@qq.com;任安虎,硕士,副教授,主要研究方向为智能交通、嵌入式系统设计等。E-mail:591277970@qq.com;白静静,硕士研究生,主要研究方向为通信与信息系统、干线交通协同优化仿真等。E-mail:1046778894@qq.com。
