【作者】 任淑云；

【导师】 毕军；

【作者基本信息】 北京交通大学 ， 交通安全工程， 2010， 硕士

【摘要】 道路交通状态估计是交通安全管理中的重要课题,也是是智能交通系统能够正常运行并发挥作用的前提和基础。交通控制和诱导系统需要对下一时刻乃至以后一些时刻的道路交通状态做出准确的估计和判断,才能做出正确的决策。只有对交通参与者做出及时的控制和诱导,才能有效地确保交通畅通和减少道路安全隐患。另外,准确的交通状态估计也是进行交通事故检测的必要条件。目前,国内外学者大多采用卡尔曼滤波和扩展卡尔曼滤波等线性方法来解决交通状态估计问题。这些方法能够在一定程度上解决道路交通状态估计问题,并且得到了广泛的应用。然而,由于交通参与者之间存在各种复杂的相互影响,使得交通行为具有高度的非线性。而卡尔曼滤波方法假设系统模型和观测模型服从线性分布,并且系统噪声和观测噪声服从高斯分布。这就使得基于卡尔曼滤波的估计方法存在很大的误差,从而表现出一定的局限性。粒子滤波是一种通过蒙特卡罗积分仿真来实现对贝叶斯滤波器递推的技术,它不做任何线性高斯假设,是一种非线性的预测方法。本文通过仿真实验对该方法进行了深入的学习和研究,并通过与线性方法EKF滤波器的对比,验证了其在解决非线性问题时的有效性。另外,针对复杂路段下传统方法的缺陷,本文将粒子滤波方法应用到处理交通状态估计问题中。通过研究二阶宏观随机交通流模型对北京城市快速路建模,并基于Matlab平台进行交通状态估计。实验结果表明,粒子滤波算法能够对道路交通流的参数指标做出较好的估计,具有良好的适用性。然而,在实验过程中发现,随着迭代次数的增加,只有少数样本具有较大的权值,一些权值较小的粒子会出现退化现象。这不但浪费了大量计算在小权值粒子上,而且也影响了样本粒子的多样化。针对这些问题,本文尝试性地将蚁群算法的思想引入到粒子滤波算法过程中,来优化粒子滤波算法的重采样过程。运用改进后的粒子滤波算法进行实例验证并与基本滤波算法进行对比,实验结果表明,该方法具有良好的准确性和鲁棒性。更多还原

【Abstract】 Traffic state estimation is an important task of Intelligent Transportation System Management & Traffic Safety, also the presupposition and basis in its operation and functioning. Either traffic control or guidance system requires accurate estimations and assessments for the road traffic state in the next or even following moments. The efficiency of the decision making of the system depends on the accuracy of traffic state estimation. Only when the traffic participants are controlled and guided efficiently, the fluency of the roads and the reduction of hidden accidents can be ensured. In addition, the accurate estimation of traffic flow is a necessary condition for traffic accident detection.Currently, most literatures adopt linear approaches such as Kalman filter and extended Kalman filter techniques to estimate the traffic state estimation. These techniques can predict traffic state in relatively simple traffic condition, which has been widely used. However, traffic behaviors have higher-order nonlinearity due to the complex interaction among participants. In this case, Kalman filter has some limitations since it assumes the both the system and observation model subject to linear distributions. Moreover, it defines both the system and observation noises obey the Gaussian distribution, which leads to significant errors for the estimation result.Particle Filter (PF) is a non-linear and non-gaussian estimation technique which realizes the recursive Bayesian Filter via Monte Carlo integration simulation. From the indepth study and investigation of PF via the simulation experiment and the comparison of the PF technique with the linear EKF, we prove that PF is efficient of solving nonlinear problem.In addition, to remedy the insufficient of the traditional methods under the complex road environment, this report proposes the usage of PF in solving the traffic state estimation problem. The model which combines PF and the second-order macroscopic random traffic flow model is established on the MATLAB Platform. Experiment results based on the real world data from Beijing Loop verify that the proposed technique can estimate the traffic state parameters acutely with good adaptability.However, PF shows some limitations such as sample poverty, particle degradation, etc. To solve these problems, the thesis innovatively proposes to apply Ant Colony Algorithm (ACO) in the update process of PF and verify the ACO-PF technique with real world data. Experiment results suggest that the proposed technique is superior to the particle filter with higher accuracy and stronger robustness.更多还原