Modelling, solution and evaluation techniques for Train Timetable Rescheduling via optimisation

Reynolds, Edwin (2021) Modelling, solution and evaluation techniques for Train Timetable Rescheduling via optimisation. PhD thesis, UNSPECIFIED.

[img]
Text (2021reynoldsphd)
2021reynoldsphd.pdf - Published Version
Available under License Creative Commons Attribution.

Download (2MB)

Abstract

It is common on railways for a single train delay to cause other trains to become delayed, multiplying the negative consequences of the original problem. However, making appropriate changes to the timetable in response to the initial delay can help to reduce the amount of further delay caused. In this thesis, we tackle the Train Timetable Rescheduling Problem (TTRP), the task of finding the best combination of timetable changes to make in any given traffic scenario. The TTRP can be formulated as an optimisation problem and solved computationally to aid the process of railway traffic control. Although this approach has received considerable research attention, the practical deployment of optimisation methods for the TTRP has hitherto been limited. In this thesis, we identify and address three outstanding research challenges that remain barriers to deployment. First, we find that existing TTRP models for large station areas are either not sufficiently realistic or cannot be solved quickly enough to be used in a real-time environment. In response, a new TTRP model is introduced that models the signalling system in station areas in fine detail. Using a new set of real instances from Doncaster station, we show that our tailored solution algorithm can obtain provably optimal or near-optimal solutions in sufficiently short times. Second, we argue that existing ways of modelling train speed in TTRP models are either unrealistic, overly complex, or lead to models that cannot be solved in real-time. To address this, innovative extensions are made to our TTRP model that allow speed to be modelled parsimoniously. Real instances for Derby station are used to demonstrate that these modelling enhancements do not incur any extra computational cost. Finally, a lack of evidence is identified concerning the fairness of TTRP models with respect to competing train operators. New evaluation techniques are developed to fill this gap, and these techniques are applied to a case study of Doncaster station. We find that unfairness is present when efficiency is maximised, and find that it mostly results from competition between a small number of operators. Moreover, we find that fairness can be improved up to a point by increasing the priority given to local trains. This work represents an important step forward in optimisation techniques for the TTRP. Our results, obtained using real instances from both Doncaster and Derby stations, add significantly to the body of evidence showing that optimisation is a viable approach for the TTRP. In the long run this will make deployment of such technology more likely.

Item Type:
Thesis (PhD)
ID Code:
152478
Deposited By:
Deposited On:
08 Mar 2021 09:30
Refereed?:
No
Published?:
Published
Last Modified:
16 Jun 2021 12:07