城市交通信号控制干道旅行时间实时预测

Real-time Link Travel Time Prediction for Urban Signalized Arterials

城市干道旅行时间预测是实时交通运营管理与交通诱导的核心问题之一,也是出行者的重要需求。文中分析了济南市经十路采集的真实数据,研究发现了交通需求和旅行时间在工作日和非工作日同时段具有较大差异、全天具有显著早晚高峰、以及工作日同时段具有相似性及波动性等特征。基于该类特性,分别改进了适用于周期性数据的卡尔曼滤波和波动性的人工神经网络2类预测模型。提出了组合预测算法,将基于历史同时段数据的卡尔曼滤波算法的预测值作为人工神经网络的输入变量,利用历史天和临近时刻的可用数据进行了预测。结果表明:在3.8km的信号控制干道上,组合预测模型平均误差低于0.9min,误差超过2min的概率低于4%,其预测性能可满足实时的交通需求。

Link travel time prediction at signalized urban arterials,as traveler＇s necessary needs,is one key module of real-time traffic operations and traffic guidance systems.Firstly,this paper analyzed the traffic fluctuation and periodicity over time in terms of field traffic volume and link travel time estimated by Jinan video surveillance systems.Then,this study presented Kalman filter and artificial neural network algorithms for travel time prediction to capture the periodicity and fluctuation of traffic flow based on available traffic data.The structure and input variables of the proposed models were determined via data analysis and tests.Finally,a novel combined model in which the prediction of KF model is regarded as one of input variables of ANN model,is developed so as to improve the prediction.By testing the used field data from Jingshi Road in Jinan with a length of 3.8km,the proposed models showed promise in improving the prediction accuracy and reliability.The average prediction error is less than 0.9minutes,and the probability of prediction errors with greater than 2minutes does not exceed a threshold of 4%.