Problems

Bilevel optimization involves optimizing two interrelated optimization tasks, where one problem (the upper level) depends on optimal solutions of another at the lower level. This part presents test problems for bilevel optimization, including single and multi-objective scenarios.

BilevelHeuristics.TestProblems.get_problemFunction
get_problem(problem)

Return F, f, bounds_ul, bounds_ll, Ψ, o where

  • F is the upper-level objective function.
  • f is the lower-level objective function.
  • bounds_ul is the upper-level bounds.
  • bounds_ll is the lower-level bounds.
  • Ψ-mapping that receives upper-level decision vector and return a Matrix containing lower-level optimal solutions.
  • o contain a sampled optimal solutions.

Example

using BilevelHeuristics
using DelimitedFiles

F, f, bounds_ul, bounds_ll, _ = TestProblems.get_problem("TP1")

algorithm = SMS_MOBO(;
        ul = SMS_EMOA(;N = 100), # upper level optimizer
        ll = NSGA2(;N = 50), # lower level optimizer
        ul_offsprings = 10,
        options_ul = Options(iterations = 100, f_calls_limit=Inf),
        options_ll = Options(iterations = 50, f_calls_limit=Inf)
    )

res = optimize(F, f, bounds_ul, bounds_ll, algorithm)

final_archive = algorithm.parameters.archive

final_archive[1][1] # upper level (solution #1)
final_archive[1][2] # lower level (solution #1)

ul_front = fvals([sol[1] for sol in final_archive])
writedlm("data.csv", ul_front, ',')
source