基于电警数据的单点自适应信号控制优化方法

Adaptive Signal Control Optimization for Isolated Intersections Based on E-police Data

摘要:

通过深入挖掘利用电警数据蕴含的交通流信息，提出了一种基于电警数据的单点自适应信号控制优化方法。研究对象为各进口道均布设电警设备的四肢交叉口，4个上游交叉口也布设有电警设备，能够获取前往目标交叉口的车流。首先，根据TOD时段上下游电警匹配历史数据，利用高斯混合模型估计各路段行程时间参数，进而标定基于截断正态分布的车队离散模型；其次，基于实时到达率预测值，以目标交叉口总延误最小为优化目标，建立整数规划模型对信号控制参数进行实时滚动优化，采用动态规划算法求解；最后，在不同需求场景下对该方法进行仿真验证。结果表明，路段行程时间均值与标准差的估计误差均小于3s，车均延误与排队长度较最优灯组配时分别减少34.2%和40.5%以上。与感应控制相比，该方法在高需求场景下改善效果更为明显，车均延误与排队长度分别减少12.9%和15.8%，在其他场景下也分别有2.6%和5.4%以上的改善。同时，滚动优化时长的增加也能带来控制效益的提升。

Abstract:

An adaptive signal control optimization method based on E-police data was proposed by deeply mining and utilizing the traffic flow information contained in E-police data. The study object is a four-arm intersection where E-police detectors are deployed at each approach. Its four upstream intersections are also equipped with E-police detectors， which can obtain the traffic flow to the target intersection. Firstly， according to the matched historical data of upstream and downstream E-police in the TOD period， the link travel time parameters were estimated by Gaussian mixture model， and then the platoon discrete model based on truncated normal distribution was calibrated； Secondly， based on the predicted value of real-time arrival rate， taking the minimum total delay of the target intersection as the optimization objective， an integer programming model was established to optimize the signal control parameters in real-time by rolling horizon scheme， which was solved by dynamic programming； Finally， the proposed methods were simulated and verified in different demand scenarios. The results show that the estimation errors of the mean and standard deviation of link travel time are less than 3s， and the average vehicle delay and queue length are reduced by more than 34.2% and 40.5% respectively compared with the optimal group-based timing. Compared with the actuated control， the improvement of the proposed method is more obvious in high demand scenarios， where the average vehicle delay and queue length are reduced by 12.9% and 15.8% respectively， and more than 2.6% and 5.4% in other scenarios respectively. At the same time， the increase of planning horizon can also improve the control benefits.