Improving real-time train dispatching: Models, algorithms and applications

Improving real-time train dispatching: Models, algorithms and applications

D'Ariano, A.

Hansen, I.A. (promotor)

Civil Engineering and Geosciences

2008-04-07

Traffic controllers monitor railway traffic sequencing train movements and setting routes with the aim of ensuring smooth train behaviour and limiting as much as existing delays. Due to the strict time limit available for computing a new timetable during operations, which so far is rather infeasible by using existing tools, railway traffic controllers usually restrict themselves mostly to a few manual timetable modifications and the chosen traffic control actions may be often sub-optimal. This PhD thesis is principally concerned with the design, implementation and evaluation of an advanced and robust laboratory tool for supporting railway traffic controllers in the everyday task of managing timetable disturbances. This dynamic traffic control system co-ordinates the speed of successive trains on open track, solves expected route conflicts and provides dynamic use of platform tracks in stations or alternative paths in a corridor between stations. Blocking time theory for modeling track occupation and signaling constraints is combined with alternative graphs for solving dynamic traffic control problems with the aim of increasing the punctuality and the use of infrastructure capacity at a network scale. The feasibility of the dispatching options is verified in a very short computation time by dynamic updating of the corresponding headways, train speeds and blocking time graphs, while the costs of the alternative dispatching options are measured in terms of maximum and average delays between consecutive trains at stations and other relevant points within the investigated network. To this end, the following achievements are included: (i) An innovative model for railway traffic optimization is presented to predict accurately train traffic flows and to enable the computation of optimal network schedules, i.e., all trains are managed simultaneously in a railway network for a given time period. (ii) The development of fast and effective scheduling algorithms based on the proposed model for the real-time management of a complex railway network is addressed. The objectives are to predict the evolution of train traffic within short computation times and to improve the punctuality by pro-actively detecting and solving train conflicts. (iii) A better use of rail capacity and a further improvement of punctuality are achieved by an iterative adjustment of train orders and routes in case of disturbances. Novel problem dedicated algorithms highlight the potential use of rerouting instead of only rescheduling the trains in order to limit the delay propagation as much as possible. (iv) Constructive algorithms for the dynamic modification of running times are provided that satisfy the timetable constraints of train orders and routes and guarantee the real-time feasibility of the running times, while respecting the signaling and safety systems in use. (v) A temporal decomposition method is introduced for the short-term traffic planning and control over a time period of up to several hours. This approach is of interest for traffic controllers since delays between running trains propagate considerably in time and space during heavily perturbed operations. (vi) A large set of computational studies on real-world instances proves that the automated decision support tool provides better solutions in terms of delay minimization compared to dispatching rules adopted by traffic controllers. Test beds are the hourly timetables of the Schiphol railway bottleneck and of the Utrecht - Den Bosch dispatching area. We study practical size instances and different types of disturbances, including multiple delayed trains, dwell time perturbations and blockage of some tracks.

real-time rail traffic management
decision support systems
alternative graph
scheduling
routing
blocking time theory
train speed coordination

http://resolver.tudelft.nl/uuid:178b886e-d6c8-4d39-be5d-03d9fa3a680f

TRAIL Research School

978-90-5584-100-4

Institutional Repository

doctoral thesis

(c) 2008 A. D'Ariano