Adaptive Operator Selection for Iterated Local Search

Abstract

Iterated local search is a simple yet powerful metaheuristic. It is only drawback is that it is quite sensitive to its only parameter: the perturbation step size. Adaptive operator selection methods are on-line adaptive algorithms that adjust the probability of applying the search operators to the current solutions. In this short note, we show the use of the adaptive pursuit algorithm to automatically select the perturbation step size for ILS when optimizing a blind, single-constraint knapsack problem. The resulting adaptive ILS achieves almost the same performance as the ILS with the best perturbation step size but without the need to determine the optimal parameter setting.