Modelling passengers' route choice behaviour on the London Underground

The purpose of this research is to compare existing route choice models with the further goal to understand whether they or the combination of them can be applied inside a transit assignment model for the London Underground (LU). A Bayesian Modelling Framework (BMF) (Fu et al., 2014) and a Random Utility Model (RUM) (Raveau et al., 2014) was applied for the same case study network. As a novel approach route choice was estimated not only for single station-to-station origin destination (OD) pairs, but stations having the same route choice properties were grouped into zones. For the BMF it was shown that for many single station–to-station OD pairs the model cannot be applied, due to the small quantity of data; but when these stations are grouped into zones that data can be sufficient and proper input for the model. It resulted that fitting a Gaussian mixture distribution of two components on the observed journey times, it gives unrealistic parameters for the components; but fitting a single Gaussian distribution, it gives appropriate values. This would suggest that all passengers take the same route, however through analysing survey data it was understood that this is not the case. This shows that when two routes have similar journey times the BMF model considers as it were one route, due to the fact that this model considers only journey time. The RUM using only the scheduled values of travel time does not consider the case when services are not running on time. It was understood that each model using only a part of the whole dataset available (only observed or only scheduled data) is like watching only a part of the whole picture. Therefore another approach was proposed which uses both dataset: Inferring the route choice from the service taken, using entry and exit times of passengers (individual Oyster data) and the actual departure and arrival times of services. It needs further research how to capture the actual times of services for the same period as Oyster data. The interdependence of crowd and route choice was discussed as the RUM uses crowd as an attribute and was not possible to determine it at this stage. Therefore it was proposed to analyse it as a transit assignment problem for a network rather than just route choice problems for single OD pairs.