Predictive Reasoning and Machine Learning for the Enhancement of Reliability in Railway Systems

Abstract

The real-time prediction of train movements in time and space is required for ensuring the reliability in operational management and in the information that is relayed to passengers. In practice, however, accurate predictions of train arrival times are very difficult to achieve, given the nature of uncertainty and unpredictability in train movements. This is often due to truly random delay causes that results in a constantly changing probability distribution in delay events as the effects of those causes. The overall consequence is less reliable estimates in train arrival times being made, which can potentially reduce the ability of traffic controllers to effectively plan and respond to disruptions. This paper presents a series of methods that are currently being applied for developing a preliminary working prototype of a future rail advisory system, which is the main objective of an ongoing PhD research project. The system prototype is expected to be capable of relaying advice to a traffic controller with the goal of minimising the effects of a disruption as much as possible and to potentially avoid future disruptions, for which accurate train movement and delay predictions using methods in predictive reasoning and machine learning are vital.