An estimation framework to quantify railway disruption parameters

Railway network operations form complex systems. Any disruption adversely impacts the operations, causing long delays. Many studies investigate the effect of a railway incident; however, a holistic quantification is lacking. This study aims to present an estimation framework for flexible traffic management systems, which can help reduce network delays and enable dispatchers to make better-informed decisions. An incident's impact on the network is estimated by creating a sequence of models, which predict two key variables. Firstly, the incident duration is predicted, which is used to predict the second variable: total delay caused by the incident. Various influencing attributes are examined, such as weather, network and railway-related attributes. Their relationship with the response variables is studied in order to understand the incident's impact. Using incident data from the Danish Railways, machine learning models are estimated. The results show that neural networks outperform other competing models for total delay modelling, resulting in improved prediction by the estimation framework, thus giving higher accuracy than the stand-alone models in the study.