滤波方法在交通状态估计中的应用研究

发布时间：2018-10-29 15:33

【摘要】：交通状态估计是实现智能交通管理的重要一环,是其能够正常运行并发挥作用的前提和关键,估计的好坏直接影响到交通控制及管理的结果。交通状态估计要解决的问题就是如何从带有随机性和不确定性的交通状态变化中,根据来自交通流信息采集设备的交通流参数数据,结合其它影响因素,进行数据的系统分析,找出其中的规律性,建立相应的预测方法和模型,以估计未来几个时段的交通状态变化趋势。交通系统的强非线性特点以及交通系统复杂度的不断增大使得传统的交通估计算法具有局限性。为了应对这一问题,本文以城市快速路为背景研究了粒子群优化粒子滤波算法及Shadowing Filter(简写为SF)算法这两种计算机优化算法在交通状态估计中的应用。在利用粒子滤波(PF)算法进行交通状态估计时,PF的重要性采样过程中采用的是次优的重要性函数。为了优化PF采样过程,本文采用将粒子群优化算法(PSO)融入PF中(称为PSOPF方法),并将其初步应用于交通流状态估计问题中。为了检验PSOPF算法的可行性及有效性,本文对PF算法和PSOPF算法分别进行了大量的数值试验,并对每次交通状态估计的精度进行了方差估计。通过实验对比发现,基于PSOPF算法的交通状态估计在精度及稳定性上明显好于PF算法。SF算法是以动力模型为约束,通过不断调整滤波窗口内的估计值与相应的前一时刻的估计值按模式向后积分一步的值之间的偏差,以滤波窗口内整体偏差最小为目标来实现状态变化轨迹的最佳估计。本文初步尝试利用SF方法获取接近观测时间序列模式轨迹的最佳估计值,并用于对交通流状态进行估计,与基于PSOPF算法的交通状态估计进行了实验对比。实验结果表明,当观测噪声大于系统噪声时,SF的性能要优于PSOPF。
[Abstract]:Traffic state estimation is an important part of intelligent traffic management. It is the premise and key of its normal operation and function. The quality of traffic state estimation directly affects the result of traffic control and management. The problem to be solved in traffic state estimation is how to analyze the data systematically according to the traffic flow parameter data collected from traffic flow information and other influencing factors from the change of traffic state with randomness and uncertainty. The regularity is found out and the corresponding prediction methods and models are established to estimate the changing trend of traffic state in the next few periods. Because of the strong nonlinearity of traffic system and the increasing complexity of traffic system, the traditional traffic estimation algorithms are limited. In order to deal with this problem, the particle swarm optimization (PSO) and Shadowing Filter (SF) algorithm are studied in this paper in the context of urban expressway. The application of particle swarm optimization (PSO) and Shadowing Filter (SF) algorithm in traffic state estimation is studied in this paper. In traffic state estimation using particle filter (PF) algorithm, the importance function of PF is used in the process of importance sampling. In order to optimize the PF sampling process, the particle swarm optimization algorithm (PSO) is incorporated into the PF (called PSOPF method) and applied to the traffic flow state estimation problem. In order to verify the feasibility and effectiveness of the PSOPF algorithm, a large number of numerical experiments have been carried out on the PF algorithm and the PSOPF algorithm, respectively, and the variance estimation of the accuracy of each traffic state estimation has been carried out. It is found that the accuracy and stability of traffic state estimation based on PSOPF algorithm is better than that of PF algorithm. SF algorithm is constrained by dynamic model. By constantly adjusting the deviation between the estimated value in the filter window and the corresponding one step value of the previous time, the optimal estimation of the state change trajectory is realized with the minimum overall deviation in the filter window as the goal. In this paper, the SF method is used to obtain the best estimation value of the pattern trajectory close to the observed time series, and it is used to estimate the traffic flow state, which is compared with the traffic state estimation based on the PSOPF algorithm. The experimental results show that when the observed noise is greater than the system noise, the performance of SF is better than that of PSOPF..
【学位授予单位】：中国海洋大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;TN713

【相似文献】