recently I've been challenged for the KEP usecase (parcel) in the following way - maybe the approach below sounds interesting to be discussed. Here's the story:
- You are given a few hundred locations where you'd like to deliver parcels
- about 200 positions in the eastern part of Karlsruhe
- As a KEP driver you would prefer to park on strategic meaningful spots from where you could handle several parcels via footwalk
- Where are meaningful spots to park?
- I use the coordinate list of the orders twice: both for the locations and the potential parkings - so in the end each recommended parking spot is at a position of an order (and we can server some more orders from over there)
- Distance Mode: I used airline but the engine can also refer to a pedestrian routing DIMA
- CoverLocationOptions I have chosen a DistanceBasedHorizon (75m - 150m - 300m) but you can also use a traveltime based approach such as (1min - 3min - 5min) depending on how much time you would need for a single parking process. Furthermore I've applied a MinimumCoverageGoal of 100% - this is needed because we want to visit all the orders
- Smaller radiusses lead to more parking spots and shorter local walking trips
- Small radius of 75m - each bar in the top right chart is a "parking manoeuver" (about 80). The height of the bar indicates the number of parcels on the temporary local trip (not more than 3)
- Larger radiusses lead to less parking spots (time consuming parking manoeuvers) but also to longer walking trips
- Large radius of 300m - about 40 parking spots with up to 10 parcels on a single local trip
- The cluster interface also offers access to scaling by using individual activities on the different order positions. Probably this is not needed for this approach
- Imagine to perform a tour optimization based on the output of those clusters. Of course you could reduce the 200 "orders challenge" to a "50-60 parking events in a row" challenge. But what is not guaranteed is that parkings in the same street have to be visited within the same transit. In some cases the tour optimization would visit one of the "parkings" on the beginning of a tour while it's "neighbour spot" is handeled at the end of the tour.
- Determine the parking spots by using the xCluster.coverLocations based on pedestrian dima.
- Perform a pedestrian sequence optimization with xTour.planTours for each parking spot and it's assigned orders to determine the "local service period"
- Perform a sequence optimization (car based DIMA) via all parking spots and use the "local service period" from step 2 as the "service period"
Best regards,
Bernd
- Here's a comparison of the results based on various radiuses: the bigger the radius is the less parking spots are returned. At the same time the average number of parcels per parking grows.
