基于虚拟动态检测的自适应信号控制方法

Adaptive Signal Control Method Based on Virtual Dynamic Detection

摘要:

鉴于传统固定检测方法难以获取连续动态车辆信息的限制，提出一种非完全网联交通环境下基于虚拟动态检测的交叉口自适应信号控制方法（ACV2D方法），以解决信号控制精度不高的问题。通过ACV2D方法构建了一个位置可变的虚拟检测断面和区间来替代传统交通流检测器，每个信号相位取得通行权后，以排队车辆中最远网联车（CV）所在位置为依据测算初始绿灯时间，同时利用测得的CV信息来预测虚拟检测断面和区间内交通流状况和绿灯持续时间，在此过程中监视虚拟检测区内交通流状况与预测结果的一致性。当预测结果出现偏差时，以车均延误最小为优化目标，构建信号控制参数的实时修正模型，并以预测的车辆到达时间为决策点，采取动态规划法以信号相位为阶段正序求解最佳信号相位配时。仿真结果表明，当CV渗透率大于50%时，ACV2D方法在中高流量下的实施效果明显优于基于强化学习的自适应信号控制方法（3DQN和3DRQN方法）。进一步研究表明，ACV2D方法的控制成效受CV渗透率和关键车道组流率比之和（Y）2个因素的共同影响；Y 值越大，确保ACV2D方法有效的CV渗透率要求就越低，反之亦然。

Abstract:

Given the limitations of traditional fixed detection methods in capturing continuous and dynamic vehicle information， we propose an adaptive signal control method （ACV2D method） based on virtual dynamic detection for intersections in a partially connected traffic environment to address the issue of low signal control accuracy. Through the ACV2D method， the position-variable virtual detection section and interval are built to replace conventional traffic flow detectors. After a signal phase gains the right of way， the initial green time is calculated based on the position of the farthest connected vehicle （CV） in the queue. Simultaneously， the measured CV data are used to predict traffic flow conditions within the virtual detection section and interval， as well as the duration of phase green time. During this process， the consistency between the predicted and actual traffic flow conditions within the virtual detection area is monitored. When the prediction results deviate， a real-time correction model for signal control parameters is constructed with the objective of minimizing the average vehicle delay. Taking the predicted vehicle arrival time as the decision point， the dynamic programming method is adopted to solve the optimal signal phase timing in a forward sequence of signal phases. Simulation results demonstrate that when the CV penetration rate exceeds 50%， ACV2D method significantly outperforms reinforcement learning-based adaptive signal control methods， such as 3DQN and 3DRQN， under medium to high traffic volumes. Further research indicates that the effectiveness of the ACV2D method is jointly influenced by two factors， i.e.， CV penetration rate and the sum of key lane group flow ratios Y. The larger the Y value， the lower the required CV penetration rate to ensure the effectiveness of the ACV2D method； conversely， the smaller the Y value， the higher the required CV penetration rate.