Most popular benchmark sets for the VRP comprise instances of several hundred
customers. To foster research in the area of very large scale vehicle routing we developed an additional set of very large scale and realistic benchmark instances. These instances are based on parcel distribution in Belgium and, therefore, resemble a real-world problem with up to 30,000 customers.
On the basis of a dataset containing deliveries in the Flemish and Brussels regions
of Belgium, we were able to extract the density of parcel demand for different zones,
i.e., the number of delivered parcels per km². We use this demand density to sample
the location of customers for one delivery day for the cities of Leuven, Antwerp, Ghent and Brussels as well as for the whole region of Flanders. For each of these geographic areas we fix the number of customers and locate them according to demand density. Each customer is assigned a random demand of either 1 (50%), 2 (30%) or 3 (20%) parcels. Furthermore, we consider two delivery scenarios.
In the first scenario, the customers are delivered by vans from a distribution center
outside of the city, where the capacity of a van is, depending on the instance, set to
100-200 parcels. In the second scenario, the distribution center is located within the
city, and customers are delivered from there via cargo-bikes or electric tricycles that
can carry 20-50 parcels at a time. Thus, the routes in the first case are rather long
(long-haul) and in the second case rather short (short-haul). We obtain 10 instances
that are diverse with respect to customer placement (more clustered versus more evenly distributed), depot location (center versus fringe), and route length (many customers versus few customers).
A more elaborate describtion of the dataset and first benchmark results can be found in the working paper. We encourage you to tackle these instances and send us your results.