Let’s say I want to divide all of the counties within a state into 4 groups and I want all of the counties in a group to be geographically contiguous. Furthermore, when I’m done, I want the total population of each of the 4 groups to be as close as possible.
I was thinking I could possibly use Clustering.jl to do the initial spatial clustering based on lat/lon of each county and then use JuMP to minimize the variance in population between groups but I don’t think that’s going to work. I’m now thinking it may be possible to simply use JuMP.jl to do this but I can’t figure out how to set the problem up.
Do the following objectives make sense?
Minimize the mean (or some other statistic) of the norms of the pairwise Haversine distance matrices for each of the 4 groups (this should lead to groups having counties that are close together) and
Minimize the standard deviation in the total populations of each of the 4 groups
If so, is this a multi-objective problem (requiring MultiJuMP.jl) or could I set it up to simply minimize the sum of the mean of the norms of the pairwise Haversine distance matrices for each of the 4 groups and the standard deviation in the total populations? Can one of the open-source solvers handle this?