How Fast-Trips Works

Dynamic Passenger Assignment(DPA) Algorithms

Fast-Trips is based on the work of Alireza Khani, Mark Hickman, and others while at University of Arizona. Associated academic papers include:

Models and solution algorithms for transit and intermodal passenger assignment (development of FAST-TrIPs model)

Hyperpaths in networks based on transit schedules

Some other useful resources include:

Teaching Materials (currently in draft form) for the Fast-Trips algorithm.

Term Glossary (currently being developed collaboratively)

System Design

One of the best ways to understand what is happening internally in Fast-Trips is to trace a single passenger through a simulation. This can be done by using the run_trace.py script found in the /scripts directory.

You can also review the following code flow chart (which is a work in progress)

Skimming

Skims are average level-of-service indicators. Given that fasttrips is trip-based and dynamic, we need to define an averaging scheme. Since fasttrips is zone-based, we do not need to worry about spatial aggregation. We are therefore left with the following dimensions:

temporal - which time range do we want to represent
trip attributes - which access/egress mode combinations and which trip purposes do we want to represent
personal attributes - user_class and value of time

Regarding 1), the user has to specify a skimming start time, a skimming end time, and a sample interval. For each sampling interval, we build a shortest path from each origin to all destinations, calculate common skims, and then average over time sampling points by taking the mean.

Regarding 2), the user has to provide values for which skims are sought. Any number of combinations can be provided as long as they are represented in the model (i.e. the access/egress mode need to be defined in the supply and the path weight file, and the purpose in the path weight file).

Regarding 3), one could provide the mean value of time of the population, or several values to create several skims.

Note that at the moment you will probably want to provide only one combination of (user_class, purpose, VoT) because of the pathfinding details mentioned in pathfinding_details.

Running skimming

Skimming can be run post-assignment by specifying “create_skims = True” in the fasttrips section of the config file. It can also be run on an unassigned network (i.e. on the service schedule as per input files) by running Run.run_fasttrips_skimming().

In both cases, fasttrips requires the user to create a csv file specifying the variables listed in the skim config file.

Implemented components

All times are in seconds, fares are in whatever currency unit is provided.

invehicle_time: The time in seconds spent inside a PT vehicle.
access_time: The time in seconds spent accessing the PT service by the specified access mode, i.e. the time to get from the origin to the first PT stop.
egress_time: The time in seconds spent egressing from the PT service by the specified egress mode, i.e. the time to get from the last PT stop to the destination.
transfer_time: The time in seconds spent transferring between two stops, necessarily by foot.
wait_time: The time in seconds spent waiting for the next service at each transfer stop. This is zero for trips without transfer; there is no wait time at the first stop, see the adaption time component.
adaption_time: FastTrips’ deterministic pathfinder uses total travel time as its objective to minimise. It finds the next service that has the earliest arrival time, and then sets the departure time of the trip such that there is no wait time at the first stop. The time (in seconds) between the start of a given skim time sampling period and the actual departure time is recorded as adaption_time.
fare: The fare in the provided currency unit.
num_transfers: The number of transfers.
gen_cost: The generalised cost calculated with the weights provided in pathweights_ft and for the specified user_class, purpose, access_mode, transit_mode, egress_mode, value_of_time combination as provided in config_ft.

Pathfinding details

FastTrips runs skimming as a post-processing step to assignment. Skimming uses deterministic pathfinding, with the only difference to the point-to-point implementation of the assignment being the built of a shortest path tree per origin (i.e. one origin to all destinations). This means the deterministic pathfinder as currently implemented determines what constitutes a shortest path and at the moment this is always with respect to time, not generalised cost.

This also means that currently, running skimming with different values of time, user classes, and purposes will not generate different paths.