25.09.17
Optimising traffic management
Source: RTM Aug/Sep 17
Taku Fujiyama, from University College London (UCL), and Lei Chen, from the University of Birmingham, on the work and early findings of the Developing and Evaluating Dynamic Optimisation for Train Control Systems project.
Capacity shortage is a major problem on the GB mainline railway. As new trains, signalling, asset and customer systems are increasingly connected to each other and produce a wealth of real-time data, it is possible to use them to better regulate and optimise railway traffic, thereby increasing the capacity.
Missing at present, however, are clever and reliable algorithms to exploit the data for traffic control, as well as fully integrated simulation environments that can test and compare different algorithms and optimisation approaches. As part of the Future Traffic Regulation and Optimisation (FuTRO) programme, the ‘Developing and Evaluating Dynamic Optimisation for Train Control Systems (DEDOTS)’ project by the UCL and the University of Birmingham provides a step towards the goal of optimised traffic management.
The core idea of UCL’s algorithm is to get trains to arrive at junctions not only at the right time, but also at the right speeds. Speed control of arriving trains is often neglected, but it can be beneficial to prevent trains from having to slow down more than necessary and then having to gain speed again when the movement authority is extended. By arriving at the right speed, junction clearance time can be shortened. This is particularly true for freight trains.
UCL’s algorithm also optimises train sequences at junctions. Currently, when two or more trains are competing for the same route, Automatic Route Setting systems can prioritise them through the conflicting section. Our system has larger control areas in which a prediction of the traffic state is made. Thus, even if there is no competition for a particular route, some trains may be better to wait for others because reordering them would be good for the future traffic state. Furthermore, our algorithms for sequence management closely work with those for train speed control – optimisation across them both allows for even better traffic management.
The developed algorithms have been thoroughly tested using the University of Birmingham’s DEDOTS simulation model. While many suppliers develop simulation programs to test their own systems, it is important to objectively compare different approaches and techniques. The University of Birmingham has capabilities for a range of post-processing analysis, allowing for comparison and benchmarking. The project has modelled the Birmingham Cross City Line and the section of the East Coast Main Line (ECML) between King’s Cross and Doncaster.
The project is now testing the DEDOTS system under various scenarios. The results so far showed that our system works better than the traditional first-come, first-served approach. In the Cross City Line simulation, where scenarios with small day-to-day perturbations were mimicked, the performance improvement, i.e. improvement in delay recovery, was limited because of the frequent stopping service. In the case of ECML, the results so far suggest potential for better delay recovery and more efficient use of capacity. Interestingly, we have found that while a larger control area leads to a better overall performance, increasing the traffic state prediction function’s time horizon does not necessarily improve the performance: in our case, a 15-minute time horizon works better than a 20-minute one. Detailed results can be seen at www.sparkrail.org or in our forthcoming journal papers.
The project is now comparing simulation results with detailed data from real train movements in collaboration with FirstGroup to further test our DEDOTS system. As universities, we hope that our systems and models will be widely used by suppliers to the GB mainline railway and beyond, so that our research eventually contributes to increased capacity and better customer experience.
The optimiser and its interface design and development are conducted by UCL (Fang Xu, Taha Ghasempour, Benjamin Heydecker, Taku Fujiyama), while simulation model development and output evaluations are the work of University of Birmingham (Lei Chen, Dave Kirkwood, Gemma Nicholson, Clive Roberts).
FOR MORE INFORMATION
To contact the University of Birmingham about its simulator, email:
E: [email protected]
Or to contact UCL about its optimiser or anything else, email:
E: [email protected]