MetaheuristicTuningBenchmark.xlsx (1.44 MB)
Metaheuristic parameter tuning dateset
dataset
posted on 2020-08-06, 11:27 authored by Ivars DzalbsIvars Dzalbs, Tatiana KalganovaTatiana KalganovaAll
thee metaheuristic algorithms (Ant Colony Optimization, Evolutionary Strategy, Imperialist Competitive Algorithm) contain multiple numerical parameters, that can
be tuned to increase the efficiency of the search. These parameters are summarised below. Candidate configurations are generated by
dividing each parameter into discrete sets - Full Fractional Design (FDD)
approach. This creates a total of 12,000 candidate configurations for ACO, 1000
for ES and 5760 for ICA, a total of 18,760 across all algorithms.
ACO: Parallel instances; Number of ants; Relative pheromone strength; relative heuristic information strength; exploitation to exploration ratio; cunning rate
ES: Population size; mutation rate; number of local iterations; swap ratio
ICA: Number of countries, imperialist ratio; number of local iterations; assimilation rate; average power of empire's colonies; independence rate
Dataset divided into three worksheets, one for each algorithm. For each configuration 10 run fitnesses are provided as well as average. With NaN representing values that did not finish within 60 seconds.