A method for classifying red signal approaches using train operational data

Yunshi Zhao, Julian Stow, Chris Harrison

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

The paper describes a novel technique to reliably classify whether or not a train approaching a signal at red actually came to a stand before the signal cleared (red aspect approaches) using the data from conventional signalling systems. This data is limited to the times at which the signal aspect changes and the times at which the train enters and leaves the signalling section ('berth') in advance of the signal. Knowing the percentage of red aspect approaches is potentially important for understanding the likelihood of a signal being passed at danger (SPAD) at individual signals and also for normalisation of SPAD data, both locally and nationally, for trending and benchmarking. The industry currently uses the number of red aspect approaches based on driver surveys as estimates, which are considered to have significant shortcomings. The development of the classification model is described together with the validation procedures. The techniques presented in this paper allow red approach rates to be reliably determined without the need for complex integration of signalling system and on-train data recorder data. The initial study of 94 million train approaches shows that 5% of them stopped at red aspects. It is also highlighted that there is a large variation in the red aspect approach rates between signalling areas and between individual signals. SPAD risk assessment at individual signals could be significantly enhanced by the ability to estimate red aspect approach rates using the techniques described.

Original languageEnglish
Pages (from-to)67-74
JournalSafety Science
Volume110
Issue numberPart B
Early online date15 Dec 2017
DOIs
Publication statusPublished - 1 Dec 2018

Fingerprint Dive into the research topics of 'A method for classifying red signal approaches using train operational data'. Together they form a unique fingerprint.

  • Cite this