Pre‐2025 public transport method
In 2025, the Planbureau voor de Leefomgeving (PBL) gave Object Vision the assignment to develop a method to not only calculate the fastest public transport routes, but also minimise travel costs. Previously, the costs were calculated for the fastest route that was calculated at that point. The quickest route was computed using Dijkstra’s algorithm on a complete routeable public transport network, including waiting at home, transfers, and waiting at stops. And adding pre- and post-transport links to the result. This worked well, since travel time is available per link. However, costs are not, as the Dutch cost model has some non-linear marginal cost components for certain operators.
To accommodate different pre- and post-transport options, multiple impedance matrices are calculated with different attached pre- and post-transport options. Such as:
| Pre | PT | Post | Description |
|---|---|---|---|
| W | PT | W | Walking -> Public transport -> Walking |
| C | PT | W | Cylcling -> Public transport -> Walking |
| C | PT | B | Cylcling -> Public transport -> Cylcling |
| C-Tic | PT | W | Cylcle to InterCity rail station -> Public transport -> Walking |
| C-TMT | PT | W | Cylcle to rail, metro, or tram stop -> Public transport -> Walking |
However, different methods should also be combined. For example, cycling to intercity stations could be interesting; however, for many areas it is not possible to cycle within a parameterised time to an InterCity station (e.g. 15 minutes). Therefore, the option to walk to a PT stop should also be included. Hence, different combinations could be defined, such as C-Tic_PT_W with W_PT_W. However, this setup could result in a traveller starting at the origin, walking to a stop and then walking from that stop towards the destination (if the origin and destination are close). In that case, we would want the traveller to walk directly to the destination. So it should always be combined with either a direct walk or a direct cycle option.
A benefit of this setup is that the space-time-dependent PT Network has relatively few turns, so the Dijkstra algorithm can calculate the fastest route relatively quickly. Moreover, by geographically tiling the origin set, it can also be calculated in parallel to fully utilise the computer’s processor and memory capacity.