Project number: COF-UOH-53 & COF-UOH-69

Using machine learning to reduce signals passed at danger

This project explored the use of machine learning techniques to significantly expand our coverage of the number of red aspect approaches across the rail network.

'The ability to extend RAATS into areas not currently covered by the S-Class data feed will be of immense value and allow wider uptake and use of the outputs from the tool.'

Chris Harrison
Chair of the RAATS user group, RSSB

The challenge

The network coverage of the Red Aspect Approaches to Signals toolkit (RAATS) is currently limited, but it would help industry if the coverage that RAATS provides was significantly increased.

That coverage is currently limited because signal aspect information (S-Class) is a necessary component of information for the algorithms that power the RAATS but are only available for part of the network.

Berth Occupancy (C-class) messages are available for the vast majority of the network, but do not provide this signal aspect information. If signal data could be inferred from the more widely available C-class data, this would enable the coverage of the RAATS tool to be significantly expanded.

What we did

Through the Strategic Partnership with the University of Huddersfield, a process was developed for cleaning data, extracting relevant features, and training a machine learning classifier. The latter could be used to provide estimates for red aspect approaches to signals with berth occupancy (C-class) data alone.

The classifier was trained on a vast set of data, taking into account berth length, berth occupancy times, train type, and network topology to make classifications.

The process and code for undertaking this approach was fully documented and will be incorporated into the upgrade of RAATSv2.

Benefits delivered

Enabling this machine learning methodology will greatly expand the coverage of the RAATS tool and enable a wider view of safety issues on the network more generally.

It will do this because the Red Approach Rates provided by RAATS are an important piece of information for safety professionals in managing signals passed at danger (SPADs). SPADs can be a precursor to train collisions and derailments, so extending the coverage of RAATS will help increase safety.