Watching the clock: can AI help with train timetabling?

Toshiba Digital and Consulting Corporation and Mitsui are providing a digital twin software package to help Greater Anglia (GA) plan UK timetabling more effectively and optimise its new fleet of rolling stock. Julian Turner gets the lowdown from GA’s head of performance and planning Keith Palmer.

Image: tonton /

Image: elisa galceran Garcia |


was in 2002, while at the University of Michigan, that Dr Michael Grieves wrote about bridging the gap between the virtual and real worlds using digital replicas of physical assets, processes or systems. Two decades on, his concept of a ‘digital twin’ has the potential to revolutionise industry.

Digital twins use sensors to gather data in real time, which is then processed in a cloud-based system before being compared with other business and contextual data. The resulting analysis enables the operator to predict problems, optimise critical processes, and drive innovation and performance.

In the context of the passenger rail industry, everything from station ticket machines and escalators to track, switches, crossings and overhead line structures can now be replicated using digital twins.

Pioneered in Japan, where it has been proven to improve punctuality, the software is now the focus of a ground-breaking project involving UK rail operator Greater Anglia (GA), and Japanese firms Toshiba Digital & Consulting Corporation (TDX) and Mitsui, which in 2017 bought a 40% stake in GA.

Having been trialled on the West Anglia route from Cambridge and Stansted Airport to London Liverpool Street, it is now being rolled-out across the network in the east of England to help Greater Anglia create and deploy more robust timetables using Toshiba’s AI-enabled Cyber Physical Systems (CPS) technology.

“The digital twin simulator model is basically a digital representation of the entire railway, with all its variances and oddities,” explains Greater Anglia head of performance and planning Keith Palmer.

“Every element that impacts or influences a railway – stations, signals, track, curves, the gradient of the lines, even rolling stock characteristics such as what trains we are running, how they accelerate, and variants in driver behaviour – is all replicated in the digital twin.

“This enables us to simulate a model at a very specific level in order to understand the impact on the overall timetable – something we cannot do, as far as I am aware, using any other current system.”

Keith Palmer – Image: Greater Anglia

The digital twin simulator model is basically a digital representation of the entire railway, with all its variances and oddities

Cultural exchange: collaborative working between GDX, GA and Mitsui

Toshiba engineers spent five months collecting data on GA’s existing timetable, the acceleration and braking performance of its rolling stock, and detailed information about the position of signals, curves and the gradient of the line, much of it courtesy of Network Rail (NR).

“The first year of the trial involved the engineers from TDX learning about how the UK rail system works, its culture, and making sure NR, as the asset owner of the track and signalling, was engaged,” says Palmer. “We had tremendous support from NR in getting access to route and signal diagrams.”

In addition to hard figures on assets and infrastructure, the TDX teams ventured out into the field in order to make themselves aware of potentially important human factors not present in the data.

Image: Diana Vucane |

“Not everything on the railway is recordable or purely data driven,” says Palmer. “Driver behaviour – how they respond to yellow signals, for example – is very difficult to model, as is the way passengers interact in or outside major stations such as Liverpool Street. It is very difficult to pull that detail out of the data, especially when you maybe don’t appreciate all the oddities of the UK rail industry.”

GA encouraged TDX to take rides in train cabs, and talk to station staff, as well as driver managers and trainers, in order to get a feel for what actually happens on the network on a day to day basis.

Powered by innovation: inside the GDX simulator

The result is a digital twin that uses AI to organise and consolidate data that the railway industry historically kept in silos – such as track infrastructure, rolling stock performance, timetables, and rule and regulations – in order to accurately reproduce a real-world train operations environment within a cyberspace.

Operated by a TDX team in Japan, the simulator alters calling patterns, train timings and platform allocations to make the timetable more efficient and robust, improving punctuality. The adjustments are checked to make sure they are feasible and, if accepted by GA train planners in the UK, inputted into the real-world timetable. The more data entered into the system, the more accurate the output.

Digital twin technology has already been proven to improve punctuality in Japan. Image: imwaltersy |

“Before now, there was nothing available that allowed us to understand the performance impact of a timetable and whether it was going to cause timing issues or conflicts – the digital twin helps us to solve that problem and understand the benefits a timetable change could bring,” explains Palmer.

“By understanding a scenario before and after it occurs, and making a better decision around the performance impact, we can develop a human understanding of the railway, passenger interaction and behaviour, and dispatch arrangements. This is not about replacing experienced train planners who develop timetables; it is about giving those staff the opportunity to do more with their time.”

The simulator  uses AI to organise and consolidate data that the railway industry historically kept in silos

The digital twin is modelling the potential impact of GA’s planned timetable rewrite in 2021

Enhanced performance: the benefits of digital twin technology

Train operation and infrastructure management became separate following rail privatisation in the UK, making creating train plans extremely complex for the simple reason that multiple companies use the same track infrastructure to provide services, an example being the West Anglia main line.

“You have airport services from Stanstead Airport to Liverpool Street that need to be fast, combined with metro-type services on inner routes that must stop frequently and provide a frequent service,” Palmer explains. “These types of services are sharing the same two-track railway with services from other operators – Cambridge, for example – so there is a real conflict of interest.

“These type of project could take a lot of planning and manpower using traditional methods, so we said to Toshiba ‘can the digital twin give us a better way of allocating station stops, for example, in order to optimise the whole route?’

“Before we could only generate one or two timetable options; now, we can produce maybe a dozen solutions and find out what impact that has commercially and operationally. This is a major step forward, and ultimately we will end up with a better timetable for the customer, which is the ultimate goal.”

Liverpool Street station in London, UK. Image: Andres Garcia Martin |

Six months into the solution’s deployment on the GA network, tangible benefits include analysis on platform optimisation at Liverpool Street, ongoing work aimed at reducing the cost of NR-imposed speed restrictions, and suggested adjustments for the impending GA timetable change in May 2020.

The digital twin is also modelling the potential impact on the railway of GA’s new fleet of rolling stock, as well as the operator’s planned timetable rewrite in 2021.

“The biggest return from the digital twin project is going to be the ability to speed up journeys, reduce running times and have better optimisation because we have new trains,” says Palmer.

“The simulator effectively takes out the old trains and replaces them with the new ones, looks at the acceleration and braking characteristics, and offers up opportunities for the future.

“The digital twin allows us to look at seven of the eight different whole timetable outputs, which can be modelled for journey times, and performance, passenger impact and commercial impact too. Thanks to our project with TDX and Mitsui, we are already looking at the bigger picture.”