Skip to main content
Operations Research

You are not logged in. Your edit will be placed in a queue until it is peer reviewed.

We welcome edits that make the post easier to understand and more valuable for readers. Because community members review edits, please try to make the post substantially better than how you found it, for example, by fixing grammar or adding additional resources and hyperlinks.

5
  • $\begingroup$ +1 for suggesting clustering. This approach is generally called cluster-first route-second. It would be great if you could add more details to improve your answer. For example, mentioning $k$-means clustering (or any other clustering approach that you would consider) as well as the related R packages in your answer would be helpful to the OP (original poster). $\endgroup$ Commented Aug 5, 2021 at 22:07
  • $\begingroup$ Hi! Thanks for the help. Yes, I tried Clustering. Specially K-means. I got the idea from the cluster first route second approach as @r.beigi suggested. The problem I have is the clusters have demand which varies a lot and it is very right-skewed. Like this: imgur.com/7qyTPrb $\endgroup$ Commented Aug 6, 2021 at 5:24
  • 1
    $\begingroup$ @stat_shib What is the number of clusters? the size of each cluster (how many demand points in each cluster)? the total demand in each cluster? the number of vehicles of each type? the capacity of each vehicle? the vector for each demand point that is used for clustering, i.e., is it just the $x$,$y$ coordinates, or it includes other features too (don't forget feature scaling if the latter is the case)? $\endgroup$ Commented Aug 6, 2021 at 7:24
  • $\begingroup$ @stat_shib You should try various heuristics based on the answers to the questions above and choose the one that works best in your special case. For example, you could set the number of clusters to the number of vehicles you have! IF you are lucky, i.e., a feasible allocation of vehicles to the clusters exist, then this simply leads to a solution (you just need to solve a TSP for each cluster)! $\endgroup$ Commented Aug 6, 2021 at 7:27
  • $\begingroup$ @r.beigi I did K-means with 200 clusters. The size and total demand of each cluster are different because the clustering is only done based on their longitude and latitude. Probably I will need to add more constraints like the total demand of each cluster with some constrained clustering. Probably you're right. I can do that and it can be a good starting solution I think. $\endgroup$ Commented Aug 6, 2021 at 9:03