Observability quantification of public transportation systems with heterogeneous data sources: An information-space projection approach based on discretized space-time network flow models
Observability quantification of public transportation systems with heterogeneous data sources: An information-space projection approach based on discretized space-time network flow models
复制标题
DOI：
10.1016/j.trb.2019.08.011
复制
发表时间：
2019-10
期刊：
影响因子：
--
通讯作者：
Jiangtao Liu;Xuesong Zhou
中科院分区：
文献类型：
--
作者：
Jiangtao Liu;Xuesong Zhou
本文着眼于如何量化不同感兴趣状态下的系统可观测性，提出了一个建模框架来系统地考虑公共交通系统中的多源传感器信息。通过开发线性方程组和不等式系统，可以根据来自异构传感器源的可用数据生成信息空间。然后，引入一些投影函数来匹配独特的信息空间与不同的系统状态之间的关系，例如，以城市轨道交通系统为研究对象，特定时间间隔的站台或车辆内的客流/密度、每个起点-目的地对的路径流量、从不同运营公司的车票中收取的收入等。通过在所提出的时空网络流模型中计算其最大可行状态范围，进一步量化由状态估计不确定性表示的相应可观测性。所有提出的模型都通过 Dantzig-Wolfe 分解作为线性规划模型来求解，并且还提出了基于 k-最短路径的近似方法来求解大规模网络中的模型。最后，进行数值实验来证明我们提出的方法和算法。
Focusing on how to quantify system observability in terms of different interested states, this paper proposes a modeling framework to systemically account for the multi-source sensor information in public transportation systems. By developing a system of linear equations and inequalities, an information space is generated based on the available data from heterogeneous sensor sources. Then, a number of projection functions are introduced to match the relation between the unique information space and different system states of interest, such as, the passenger flow/density on the platform or in the vehicle at specific time intervals, the path flow of each origin-destination pair, the earning collected from the tickets to different operation companies etc., in urban rail transit systems as our study object. Their corresponding observability represented by state estimate uncertainties is further quantified by calculating its maximum feasible state range in proposed space-time network flow models. All of proposed models are solved as linear programming models by Dantzig–Wolfe decomposition, and a k-shortest-path-based approximation approach is also proposed to solve our models in large-scale networks. Finally, numerical experiments are conducted to demonstrate our proposed methodology and algorithms.