Train delay predictions based on Bayesian networks including inter-train conflict dependencies


METADATA ONLY
Loading...

Date

2022-05-18

Publication Type

Other Conference Item

ETH Bibliography

yes

Citations

Altmetric
METADATA ONLY

Data

Rights / License

Abstract

Train delays are inconvenient for passengers. Due to numerous influencing factors, delays are unavoidable in railways and to minimize their cascading consequences, traffic control managers aim to avoid their spreading. As of the limited track infrastructure, this task is difficult with often very restricted options. Therefore, it is important to provide real- time predictions of train delays that estimate the consequences of actual delays and potential traffic control actions. Recently, Bayesian networks (BN) have become a popular model to predict train delays in real-time as combination of data-driven models and operation modelling approaches. One important step of BN models consists of assuming a dependency structure of train events (arrival, departure and through) on a station level for the prediction horizon in form of a directed acyclic graph (DAG). This DAG is most widely modelled on the Markov property assumption for stochastic processes (1st and 2nd -order) of delays along train journeys. In this study, we analyze the benefits of including inter-train dependencies in a test-case setting within a corridor of the Swiss railway network and can show an increase of the prediction accuracy of 19.8% in terms of mean absolute error and 30.6% of root mean squared error.

Permanent link

Publication status

published

External links

Editor

Book title

Journal / series

Volume

Pages / Article No.

Publisher

STRC

Event

22nd Swiss Transport Research Conference (STRC 2022)

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Railways; Train delay; Prediction; Bayesian networks; Dependency structure

Organisational unit

09611 - Corman, Francesco / Corman, Francesco check_circle
02655 - Netzwerk Stadt u. Landschaft ARCH u BAUG / Network City and Landscape ARCH and BAUG

Notes

Conference lecture held on May 18, 2022.

Funding

Related publications and datasets