station-demand-forecasting-tool/sdft

R package for the Station Demand Forecasting Tool

Station Demand Forecasting Tool

This is the R package for the Station Demand Forecasting Tool.

About the tool

This tool generates a rail demand forecast (predicted trips per year) for one or more proposed local railway stations in mainland GB. If required it can also produce an analysis of potential abstraction of journeys from existing stations, enabling the net impact of a new station on rail demand to be estimated. Forecasts for multiple stations can be accommodated as part of the same job. These can be treated independently (alternative station locations are to be assessed) or concurrently (the proposed stations will coexist).

The underlying model is based on research by Marcus Young at the University of Southampton’s Transportation Research Group. At its core is a trip end model which has been calibrated on the smaller stations (network Rail Categories E and F) in Great Britain. In such a model the number of trips is a function of the population in a station’s catchment and a range of other variables, such as service frequency and number of jobs nearby. A novel aspect of this model is that probability-based catchments are defined at the unit postcode level using a station choice model. Rather than assuming everyone will use their nearest station, this provides a more realistic representation of behaviour and allows competition to occur between stations.

A conference paper with more details about the model is available. Note that the web front-end referred to in this paper is not part of the code release.

Tool implementation

While this R package controls the tool, it is not standalone. Most of the heavy lifting takes place in a PostgreSQL database that requires setting up with data tables and additional wrapper functions for pgRouting.

A Docker implementation is available.

Tool documentation

https://www.stationdemand.org.uk.

How to cite

Please cite sdft if you use it. Get citation information using: citation(package = 'sdft'):

To cite the sdft package in publications, please use the following. You can obtain
the DOI for a specific version from: https://zenodo.org/record/4066924

  Marcus Young and Simon Blainey (2020). sdft: The Station Demand Forecasting Tool.
  R package version 0.3.1. https://doi.org/10.5281/zenodo.4066924

A BibTeX entry for LaTeX users is

  @Manual{,
    author = {{Marcus Young} and {Simon Blainey}},
    title = {{sdft: The Station Demand Forecasting Tool}},
    year = {2020},
    note = {{R package version 0.3.1}},
    doi = {{10.5281/zenodo.4066924}},
  }

sdft's Issues

See Project tab for general project todos

Frequency group adjustments

Should frequency group adjustment be made in the before catchment in the abstraction analysis? Maybe it's OK to include it so isolating just the effect of the new stations?

Capture situation where no rows are returned during abstraction analysis

For example, Llanwern with abstraction station RUA. This was a mistake as RUA is no where near Llanwern. But the AFTER choiceset returned empty and error thrown:

1. Need to trap this error
2. Not clear why the AFTER should return an empty choicset dataframe in this scenario. May be a bug or bad logic in how the AFTER choiceset is generated.

Create station catchments returning GeoJSON

Workplace service areas in output

Add the 2-minute workplace service areas to the output in GeoJSON format.

Station distance bug

From the Llanwern example it looks like the distances are being measured from the station location NGR and not the access point NGR. It's not obvious where this is happening.

Modify trip end model so that second stage of the two stage decay function resets distance to zero and use the new calibrated model based on this modification.

Backcasting

Add the ability for the tool to automatically carry out a backcast (forecast demand for a station that opened since the calibration year).

This will require the station(s) that the backcast is for, and any stations that have been opened since, being temporarily removed from the stations table and the centroidnodes table.

Deal with error trap when no distance returned for a station in choiceset

In sdr_generate_choicesets() I have an error code of -9999 if no distance is returned for a station in a choiceset. Need to decide what action to take in this instance and code.

Need to think about what checks on inputs are needed over and above checks that will be made on the front end.

openroads - guided busway

Just check how this was dealt with?

Exogenous input bug

Seems to be an issue with the population column in the exogneous table not being populated when type is houses (and possibly when population). Appears in example model for Llanwern.

Workplace zone centroids

How can we be sure that zones within 1 minute actually fall within the 1 minute service area? The centroid may be off the road network? Should the centroid be moved to nearest edge?