The measurements on actual traffic have revealed the existence of meta-stable states with high flow. Such nonlinear phenomena have not been observed in the classic Nagel-Schreckenberg traffic flow model. Here we just ...The measurements on actual traffic have revealed the existence of meta-stable states with high flow. Such nonlinear phenomena have not been observed in the classic Nagel-Schreckenberg traffic flow model. Here we just add a constraint to the classic model by introducing a velocity-dependent randomization. Two typical randomization strategies are adopted in this paper. It is shown that the Matthew effect is a necessary condition to induce traffic meta-stable states, thus shedding a light on the prerequisites for the emergence of hysteresis loop in the fundamental diagrams.展开更多
On-line estimation of the state of traffic based on data sampled by electronic detectors is important for intelligent traffic management and control. Because a nonlinear feature exists in the traffic state, and becaus...On-line estimation of the state of traffic based on data sampled by electronic detectors is important for intelligent traffic management and control. Because a nonlinear feature exists in the traffic state, and because particle filters have good characteristics when it comes to solving the nonlinear problem, a genetic resampling particle filter is proposed to estimate the state of freeway traffic. In this paper, a freeway section of the northern third ring road in the city of Beijing in China is considered as the experimental object. By analysing the traffic-state characteristics of the freeway, the traffic is modeled based on the second-order validated macroscopic traffic flow model. In order to solve the particle degeneration issue in the performance of the particle filter, a genetic mechanism is introduced into the resampling process. The realization of a genetic particle filter for freeway traffic-state estimation is discussed in detail, and the filter estimation performance is validated and evaluated by the achieved experimental data.展开更多
Traffic flow prediction has been applied into many wireless communication applications(e.g., smart city, Internet of Things). With the development of wireless communication technologies and artificial intelligence, ho...Traffic flow prediction has been applied into many wireless communication applications(e.g., smart city, Internet of Things). With the development of wireless communication technologies and artificial intelligence, how to design a system for real-time traffic flow prediction and receive high accuracy of prediction are urgent problems for both researchers and equipment suppliers. This paper presents a novel real-time system for traffic flow prediction. Different from the single algorithm for traffic flow prediction, our novel system firstly utilizes dynamic time wrapping to judge whether traffic flow data has regularity,realizing traffic flow data classification. After traffic flow data classification, we respectively make use of XGBoost and wavelet transform-echo state network to predict traffic flow data according to their regularity. Moreover, in order to realize real-time classification and prediction, we apply Spark/Hadoop computing platform to process large amounts of traffic data. Numerical results show that the proposed novel system has better performance and higher accuracy than other schemes.展开更多
Taking into account the startup behaviour of following vehicles in the waiting area of urban traffic signal, we propose an extended Nagel-Schreckenberg model for single-lane traffic flow, in which the dynamic behavior...Taking into account the startup behaviour of following vehicles in the waiting area of urban traffic signal, we propose an extended Nagel-Schreckenberg model for single-lane traffic flow, in which the dynamic behavior of each vehicle depends on not only its own headway but also the headway of the immediately preceding one. The numerical simulation of the present model reproduces some complicated nonlinear phenomena observed in real traffic such as free flow, ghostly blockage, synchronized flow and so on. For specific parameter combinations the flow-density relation of this model shows two meta-stable branches near the transition density from free flow to wide moving traffic jam. Finally, the analytical results of the model under some specific parameters are given by using the mean field theory.展开更多
物理信息深度学习(physics-informed deep learning, PIDL)是一种将深度学习与物理学先验知识相结合的新兴范式,该范式在智能交通领域,尤其在交通状态估计应用中,展现出了巨大潜力。为进一步优化物理信息深度学习模型在交通状态估计问...物理信息深度学习(physics-informed deep learning, PIDL)是一种将深度学习与物理学先验知识相结合的新兴范式,该范式在智能交通领域,尤其在交通状态估计应用中,展现出了巨大潜力。为进一步优化物理信息深度学习模型在交通状态估计问题上的准确度与收敛速度,构建了一个结合Aw-Rascle宏观交通流模型的物理信息自适应深度学习模型(physics-informed adaptive deep learning with Aw-Rascle, PIAdapDL-AR),依据有限与局部的交通检测数据,实时准确估计全局交通流状态。主要的改进包括两部分,一是在PIDL框架中的物理部分引入高阶Aw-Rascle交通流模型作为物理约束条件,引导并规范神经网络的训练过程;二是在神经网络部分融合自适应激活函数,替代固定的非线性激活函数,以动态优化神经网络性能。基于NGSIM数据集生成模拟的固定检测器数据和移动检测器数据,进行实验以验证模型有效性。实验结果表明:在不同覆盖率的固定检测数据场景下,PIAdapDL-AR的相对误差相比于基线模型PIDL-LWR降低了34.38%~45.24%;在不同渗透率的移动检测数据场景下,PIAdapDL-AR的相对误差相比于PIDL-LWR降低了18.33%~34.95%;融合自适应激活函数的PIAdapDL-AR的收敛速度优于配置固定激活函数的PIDL-AR,且收敛速度和估计精度均随着自适应激活函数中比例因子的增大而提升。展开更多
交通流预测是交通智慧化的重要任务。为了提高短时交通流的预测精度,解决单一模型预测精度不足、易受噪声干扰等特点,提出了一种基于随机拓扑优化的提升回声状态网络(echo state networks, ESN)的预测方法。该方法以简单的回声状态网络...交通流预测是交通智慧化的重要任务。为了提高短时交通流的预测精度,解决单一模型预测精度不足、易受噪声干扰等特点,提出了一种基于随机拓扑优化的提升回声状态网络(echo state networks, ESN)的预测方法。该方法以简单的回声状态网络为基本构成单元,利用随机拓扑优化策略对回声状态网络的拓扑结构进行优化选择,然后利用基于误差补偿的提升算法提高整体模型的预测精度。通过对随机拓扑优化策略和提升算法在实际交通流预测问题中的性能分析,验证了所提方法的可行性和有效性,同时可为其他弱预测学习器的学习性能改进提供参考。展开更多
文摘The measurements on actual traffic have revealed the existence of meta-stable states with high flow. Such nonlinear phenomena have not been observed in the classic Nagel-Schreckenberg traffic flow model. Here we just add a constraint to the classic model by introducing a velocity-dependent randomization. Two typical randomization strategies are adopted in this paper. It is shown that the Matthew effect is a necessary condition to induce traffic meta-stable states, thus shedding a light on the prerequisites for the emergence of hysteresis loop in the fundamental diagrams.
基金Project supported by the National High Technology Research and Development Program of China (Grant No. 2011AA110303)
文摘On-line estimation of the state of traffic based on data sampled by electronic detectors is important for intelligent traffic management and control. Because a nonlinear feature exists in the traffic state, and because particle filters have good characteristics when it comes to solving the nonlinear problem, a genetic resampling particle filter is proposed to estimate the state of freeway traffic. In this paper, a freeway section of the northern third ring road in the city of Beijing in China is considered as the experimental object. By analysing the traffic-state characteristics of the freeway, the traffic is modeled based on the second-order validated macroscopic traffic flow model. In order to solve the particle degeneration issue in the performance of the particle filter, a genetic mechanism is introduced into the resampling process. The realization of a genetic particle filter for freeway traffic-state estimation is discussed in detail, and the filter estimation performance is validated and evaluated by the achieved experimental data.
基金partly supported by the National Natural Science Foundation of China(Grants No.61571240,61671474)the Jiangsu Science Fund for Excellent Young Scholars(No.BK20170089)+2 种基金the ZTE program“The Prediction of Wireline Network Malfunction and Traffic Based on Big Data,”(No.2016ZTE04-07)Postgraduate Research&Practice Innovation Program of Jiangsu Province(No.KYCX18_0916)the Priority Academic Program Development of Jiangsu Higher Education Institutions
文摘Traffic flow prediction has been applied into many wireless communication applications(e.g., smart city, Internet of Things). With the development of wireless communication technologies and artificial intelligence, how to design a system for real-time traffic flow prediction and receive high accuracy of prediction are urgent problems for both researchers and equipment suppliers. This paper presents a novel real-time system for traffic flow prediction. Different from the single algorithm for traffic flow prediction, our novel system firstly utilizes dynamic time wrapping to judge whether traffic flow data has regularity,realizing traffic flow data classification. After traffic flow data classification, we respectively make use of XGBoost and wavelet transform-echo state network to predict traffic flow data according to their regularity. Moreover, in order to realize real-time classification and prediction, we apply Spark/Hadoop computing platform to process large amounts of traffic data. Numerical results show that the proposed novel system has better performance and higher accuracy than other schemes.
文摘Taking into account the startup behaviour of following vehicles in the waiting area of urban traffic signal, we propose an extended Nagel-Schreckenberg model for single-lane traffic flow, in which the dynamic behavior of each vehicle depends on not only its own headway but also the headway of the immediately preceding one. The numerical simulation of the present model reproduces some complicated nonlinear phenomena observed in real traffic such as free flow, ghostly blockage, synchronized flow and so on. For specific parameter combinations the flow-density relation of this model shows two meta-stable branches near the transition density from free flow to wide moving traffic jam. Finally, the analytical results of the model under some specific parameters are given by using the mean field theory.
文摘物理信息深度学习(physics-informed deep learning, PIDL)是一种将深度学习与物理学先验知识相结合的新兴范式,该范式在智能交通领域,尤其在交通状态估计应用中,展现出了巨大潜力。为进一步优化物理信息深度学习模型在交通状态估计问题上的准确度与收敛速度,构建了一个结合Aw-Rascle宏观交通流模型的物理信息自适应深度学习模型(physics-informed adaptive deep learning with Aw-Rascle, PIAdapDL-AR),依据有限与局部的交通检测数据,实时准确估计全局交通流状态。主要的改进包括两部分,一是在PIDL框架中的物理部分引入高阶Aw-Rascle交通流模型作为物理约束条件,引导并规范神经网络的训练过程;二是在神经网络部分融合自适应激活函数,替代固定的非线性激活函数,以动态优化神经网络性能。基于NGSIM数据集生成模拟的固定检测器数据和移动检测器数据,进行实验以验证模型有效性。实验结果表明:在不同覆盖率的固定检测数据场景下,PIAdapDL-AR的相对误差相比于基线模型PIDL-LWR降低了34.38%~45.24%;在不同渗透率的移动检测数据场景下,PIAdapDL-AR的相对误差相比于PIDL-LWR降低了18.33%~34.95%;融合自适应激活函数的PIAdapDL-AR的收敛速度优于配置固定激活函数的PIDL-AR,且收敛速度和估计精度均随着自适应激活函数中比例因子的增大而提升。
文摘交通流预测是交通智慧化的重要任务。为了提高短时交通流的预测精度,解决单一模型预测精度不足、易受噪声干扰等特点,提出了一种基于随机拓扑优化的提升回声状态网络(echo state networks, ESN)的预测方法。该方法以简单的回声状态网络为基本构成单元,利用随机拓扑优化策略对回声状态网络的拓扑结构进行优化选择,然后利用基于误差补偿的提升算法提高整体模型的预测精度。通过对随机拓扑优化策略和提升算法在实际交通流预测问题中的性能分析,验证了所提方法的可行性和有效性,同时可为其他弱预测学习器的学习性能改进提供参考。
