Modelling Public Transport Route Choice with Multiple Access and Egress Modes

Modelling Public Transport Route Choice with Multiple Access and Egress Modes


Ties Brands, Goudappel Coffeng / University Of Twente, Erik De Romph, Omnitrans International / TU Delft, Tim Veitch, Veitch Lister Consulting


To correctly assess effects of policy measures, detailed modelling of public transport trips is essential. Therefore, this route choice model contains multiple access and egress modes and multiple routing, while computation time is still limited.


The current traffic system faces well known problems like congestion, environmental impact and use of public space. Public transport (PT) is an important mode to alleviate these problems. To be able to assess the effects of policy measures properly, it is important to model the behaviour of the (public transport) traveller in a realistic way. One aspect that lacks realism in a lot of current models is the rigid separation between modes. Within many models a traveller cannot choose to switch between modes, so multimodal trips that combine a public transport trip with car or bicycle are not (or at least not explicitly) taken into account. The use of the bicycle as an access mode is very popular in the Netherlands, and becoming more and more popular in other countries too. With bike rental systems popping up in cities over the world, the bicycle becomes interesting as an egress mode as well. The use of the car as an access mode is very popular in the US and Australia. Furthermore, different users have different preferences (i.e. a fast route or a route without a transfer), which can be achieved by adding multiple routing. These two aspects together will result in more realistic transit modelling.

Multiple routing is generally achieved in two ways: stochastic assignment (draw link attributes or choice parameter values from a distribution and search for shortest paths in the modified network multiple times) or based on optimal strategies (take the departure time of vehicles into account in order to determine whether a choice option is the shortest route for some moment in time). Furthermore, it is possible to introduce crowding, which results in an iterative equilibrium assignment. Finally, it is possible to do a dynamic assignment, taking the complete schedule into account, which is very computationally expensive.

In this paper we describe a static route choice algorithm with multiple routing and multiple access and egress modes, that is capable to assess large scale networks. The route choice algorithm is multi-class, with each class corresponding to an access & egress mode pair (e.g. bicycle access, walk egress). The route choice model, which is applied identically for each class, employs a nested logit structure (with at most 4 levels) to assign demand to competing alternatives. The levels relate to the choice of access stop group, access stop, egress stop group and transit line. “Stop groups” are user defined, and can be employed to group together stops with similar attributes (e.g. bus stops, or train stations). This is important in networks containing, for example, densely located bus stops which compete with sparse train stations. The algorithm builds paths backwards, and uses highly configurable “stop finding” criteria to provide the user with appropriate control of the choice set, as well to minimize computation time.

We applied this algorithm in a real world case study in the Amsterdam metropolitan area. Multiple access and egress modes are incorporated in the mode choice process, resulting in several mode chains. These mode chains include single mode travel options (like the car or PT combined with walking), but also multimodal travel options (that always include a PT leg, in our case bicycle – PT – walk, car – PT – walk, walk – PT – bicycle, walk – PT – car). Based on the route choice model described earlier, cost matrices are calculated for each mode chain. Based on these cost matrices, a nested logit model is applied to determine mode choice, which has two nests: car in one nest and all other travel options that contain PT as a main mode in the other nest. Given the total OD demand, this results in an OD matrix for all modes and mode chains. Finally, these OD matrices are assigned to the network, again using the route choice algorithm described above. This case shows that the computation times are reasonable, the results are plausible and conceptually sound.

Our approach enables modellers to assess infrastructural network developments in large scale networks, taking into account realistic behaviour of travellers, namely the combination of multiple modes and multiple routes to reach their destination.


Association for European Transport