Link-based measurement model to estimate route choice parameters in urban pedestrian networks
Link-based measurement model to estimate route choice parameters in urban pedestrian networks
复制标题
DOI：
10.1016/j.trc.2018.05.013
复制
发表时间：
2018-08
影响因子：
8.3
通讯作者：
Yuki Oyama;E. Hato
中科院分区：
文献类型：
--
作者：
Yuki Oyama;E. Hato
利用全球定位系统（GPS）的被动监测越来越多地用于自动监测行程数据。然而，GPS跟踪数据包括测量误差，其取决于模型中的监控设备和网络描述。在城市步行网络的情况下，如城市中心，街道的建成环境往往是多样化的，这对测量有很大的影响。这些误差会导致路径选择的观测值有偏，从而导致路径选择模型的参数估计结果也有偏。为了处理这个问题的有偏估计，本研究提出了一个基于链接的路线测量模型，依次推断链接使用分解序列的数据和估计的链接特定的方差的GPS测量误差。我们还结合了一个基于链路的路由选择模型作为先验知识，通过考虑行为机制来修正测量模型，而不需要路径枚举。此外，为了消除先验信息中包含的偏差，本研究提出了一种结构估计方法，其中行为参数的不动点问题通过迭代过程来解决。通过一个数值例子和一个真实的行人网络的案例研究，所提出的方法的性能进行了检查。作为结果，该方法改进了路线测量模型的性能，并且通过结构估计方法获得的路线选择模型的估计参数与使用有偏路线选择观测的参数相比偏差较小并且表现出不同的趋势。此外，GPS测量误差的估计方差是现实的。
Passive monitoring with the Global Positioning System (GPS) is increasingly used to automatically monitor trip data. However, GPS tracking data includes measurement errors that depend on the monitoring device and network description in the model. In the case of urban pedestrian networks, such as city centers, the built environment of streets is often diverse, and this has a significant impact on the measurement. The errors cause the biased observations of route choices, and thus the parameter estimation results of route choice models are also biased. To deal with this problem of biased estimation, this study proposes a link-based route measurement model that sequentially infers links using decomposed sequences of data and estimates the link-specific variance of the GPS measurement error. We also incorporate a link-based route choice model as the prior to correct the measurement model by considering behavioral mechanism without path enumeration. Additionally, to remove the biases included in the prior information, this study proposes a structural estimation method in which the fixed point problem of behavioral parameter is solved by the iteration process. The performance of the proposed methods is examined both through a numerical example and a case study on a real pedestrian network. As the results, the methods refine the performance of the route measurement model, and the estimated parameters of a route choice model obtained by the structural estimation method are less biased and exhibit a different trend than those using the biased route choice observations. Also, the estimated variances of the GPS measurement errors are realistic.