Disruption analytics in urban metro systems with large-scale automated data
File(s)
Author(s)
Zhang, Nan
Type
Thesis
Abstract
Urban metro systems are frequently affected by disruptions such as infrastructure malfunctions, rolling stock breakdowns and accidents. Such disruptions give rise to delays, congestion and inconvenience for public transport users, which in turn, lead to a wider range of negative impacts on the social economy and wellbeing. This PhD thesis aims to improve our understanding of disruption impacts and improve the ability of metro operators to detect and manage disruptions by using large-scale automated data.
The crucial precondition of any disruption analytics is to have accurate information about the location, occurrence time, duration and propagation of disruptions. In pursuit of this goal, the thesis develops statistical models to detect disruptions via deviations in trains’ headways relative to their regular services. Our method is a unique contribution in the sense that it is based on automated vehicle location data (data-driven) and the probabilistic framework is effective to detect any type of service interruptions, including minor delays that last just a few minutes. As an important research outcome, the thesis delivers novel analyses of the propagation progress of disruptions along metro lines, thus enabling us to distinguish primary and secondary disruptions as well as recovery interventions performed by operators.
The other part of the thesis provides new insights for quantifying disruption impacts and measuring metro vulnerability. One of our key messages is that in metro systems there are factors influencing both the occurrence of disruptions and their outcomes. With such confounding factors, we show that causal inference is a powerful tool to estimate unbiased impacts on passenger demand and journey time, which is also capable of quantifying the spatial-temporal propagation of disruption impacts within metro networks. The causal inference approaches are applied to empirical studies based on the Hong Kong Mass Transit Railway (MTR). Our conclusions can assist researchers and practitioners in two applications: (i) the evaluation of metro performance such as service reliability, system vulnerability and resilience, and (ii) the management of future disruptions.
The crucial precondition of any disruption analytics is to have accurate information about the location, occurrence time, duration and propagation of disruptions. In pursuit of this goal, the thesis develops statistical models to detect disruptions via deviations in trains’ headways relative to their regular services. Our method is a unique contribution in the sense that it is based on automated vehicle location data (data-driven) and the probabilistic framework is effective to detect any type of service interruptions, including minor delays that last just a few minutes. As an important research outcome, the thesis delivers novel analyses of the propagation progress of disruptions along metro lines, thus enabling us to distinguish primary and secondary disruptions as well as recovery interventions performed by operators.
The other part of the thesis provides new insights for quantifying disruption impacts and measuring metro vulnerability. One of our key messages is that in metro systems there are factors influencing both the occurrence of disruptions and their outcomes. With such confounding factors, we show that causal inference is a powerful tool to estimate unbiased impacts on passenger demand and journey time, which is also capable of quantifying the spatial-temporal propagation of disruption impacts within metro networks. The causal inference approaches are applied to empirical studies based on the Hong Kong Mass Transit Railway (MTR). Our conclusions can assist researchers and practitioners in two applications: (i) the evaluation of metro performance such as service reliability, system vulnerability and resilience, and (ii) the management of future disruptions.
Version
Open Access
Date Issued
2021-10
Date Awarded
2022-03
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Anderson, Richard
Graham, Daniel
Publisher Department
Civil and Environmental Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)