The present model uses a synthetic population to generate the daily parcel demand of a region. While it is based on marginal survey data from France, it can theoretically be applied to other use cases as well.

The basic idea is to take a list of synthetic (or real) households and persons including their sociodemographic attributes. Those should include household size, socioprofessional category (following the French defintiion), and age per person.

After, an Iterative Proportional Fitting procedure is applied to the data to attach an average number of parcels per year to each household, based on the sociodemographic attributes. This value is then transformed into an average number of parcels per day for each household, which, in a final step, is transformed into a discrete value using a Poisson sampling process. The methodology is described in detail in

Hörl, S., Puchinger, J., 2022. From synthetic population to parcel demand: A modeling pipeline and case study for last-mile deliveries in Lyon. Paper accepted for presentation at the Transport Research Arena (TRA) 2022, November 2022, Lisbon.

The model is packaged as a Jupyter notebook. All dependencies to run the model have been collected in a conda environment, which is available in the LEAD repository as environment.yml:

conda env create -n parcels -f environment.yml

To run the model, a synthetic population (for instance, as created by the IRTX population synthesis model), needs to be located in an arbitrary directory denoted by /irtx-synpop/output. The structure of this directory should look as follows:

/irtx-synpop/output/lead_homes.gpkg
/irtx-synpop/output/lead_persons.csv
/irtx-synpop/output/lead_activities.csv

The output of the model is one geographic file that contains for each household with at least one parcel that has been generated for the synthetic day:

• Coordinates (location) of the household
• Number of parcels to be delivered
• Identifier of the household to link back to the synthetic population

Assuming that /irtx-parcels/output has been defined as the output path of the model (see below), the resulting file will be created as /irtx-parcels/output/lead_parcels.gpkg.

To run the model, the conda environment needs to be prepared and entered. After, the model is packaged as a jupyter notebook which can be run programmatically using papermill, which is part of the environment dependencies. Including the full set of command line options, the model can be run as follows:

papermill "Generate Parcels.ipynb" /dev/null \
  -pinput_path /irtx-synpop/output \
  -poutput_path /irtx-parcels/output \
  -pinput_prefix lead_ \
  -poutput_prefix lead_ \
  -prandom_seed 0 \
  -pscaling 1.0

The mandatory parameters are detailed in the following table:

The following technical parameters are available:

Using these parameters, one can, for instance, process xyz_persons.csv and write abc_parcels.gpkg in the same directory, depending on which population scenarios should be read and which parcel scenario should be written.

Finally, scenario parameters exist that can be configured:

For the Lyon living lab, some standard scenarios can be run:

• Using an unscaled output from the population synthesis model to create the baseline parcel demand for 2022.
• Using an unscaled output from the population synthesis model to create the parcel demand for 2030, with an increase of parcel by a factor of two.

The scaling factor could also be an input to create multiple different scenarios on the LEAD platform.

Baseline demand 2022

-pinput_path /irtx-synpop/output \
-poutput_path output \
-pinput_prefix lead_2022_100pct_ \
-poutput_prefix lead_2022_ \
-pscaling 1.0

Future demand 2030

-pinput_path /irtx-synpop/output \
-poutput_path output \
-pinput_prefix lead_2030_100pct_ \
-poutput_prefix lead_2030_ \
-pscaling 2.0