Portsmouth and Southsea station (Photo Network Rail)

UK: South Western Railway and the University of Portsmouth have co-operated to develop machine learning software which analyses real-time train running data to automatically identify services at risk of delay and assist controllers with selecting and implementing the most appropriate contingency plan.

The creation of a knowledge transfer partnership to invest in technology development is a requirement of First MTR’s contract to operate SWR services, and the machine learning concept was selected as a suitable idea. Innovate UK has covered 50% of the project cost.

SWR Class 458

The machine learning software makes use of data which is available to SWR as a result of FirstGroup’s ongoing project to develop an Amazon Web Services data hub for its businesses. This brings together a wide range of raw information generated by both train operators and Network Rail, and converts it into usable feeds which can be accessed by APIs. Sources include the TRUST system which records train running data, as well as the more precise location data available from onboard GPS units.

The University of Portsmouth team specialises in using machine learning to meet the needs of outside companies. The team spent a lot of time with SWR controllers to understand their needs, how railway tasks are undertaken and how ML could enable control room staff to make better informed decisions.

The software ‘enables controllers to do what they are best at’, senior research fellow at the Innovative Industrial Research group and knowledge transfer partnership academic supervisor Dr Edward Smart told Rail Business UK. ‘They understand the trains.’

The process has three stages. In the first step, the tool analyses the TRUST and GPS data to detect automatically any trains that are behind schedule by more than a minimum threshold, currently set at 2 min.

London Waterloo station (Photo: Network Rail)

In the second stage, it identifies the likely impact on other services nearby — ‘a train on a branch line might be less of a problem than on the main line at Basingstoke’, Smart explained — and also looks at the potential knock-on effects on rolling stock and crew positioning. This analysis can be generated in a matter of seconds, replacing manual and spreadsheet-based processes.

In the third stage, Smart explained, the controllers can identify and implement the most appropriate contingency plan, with ‘all data at their fingertips’.

The tool can also be used to drive performance improvement, using past data to assess the impact of different approaches to problems that have arisen. This enables the operator to revise its contingency plans to be more effective.

The software is also able to identify potential conflicts between trains. Chris Prior, Head of Control Projects at SWR, said this had proved particularly useful on the Windsor Lines, where trains run in quick succession and decisions must be made about the optimal order to path them through junctions.

Data-driven projects

The tool has been running on laptops at the South Western Railway Control Centre for around two months. The next step in the project will be to produce a production version.

FirstGroup said there may also be opportunities to roll out the system at its TransPennine Express business, which faces similar operational challenges to SWR.

London Waterloo (Photo: Network Rail)

Smart told Rail Business UK that he had met some ‘incredible individuals’ in the rail industry during the project. Providing them with access to structured and reliable data could enable a ‘whole host’ of projects to improve operational performance, he believed.

Prior suggested that there could be future opportunities to tighten up train timings, with real-time data available to the second rather than the ½ min that railways have traditionally used for timetable planning.

SWR logo on train

The project has also highlighted where there are gaps or inconsistencies in available industry data, which will enable FirstGroup and its partners to seek improvements.

‘Big data is clearly the way forward for a more proactive approach to control’, said Prior, adding that the use of AI to search through real-time data ‘takes a lot of human effort out, and makes it quicker to solve problems.’

  • SWR’s work with the University of Portsmouth has led to a wider working relationship, including a modern apprenticeship programme run by the university for the operator.