An ILS-BanditVND Approach to the Team Orienteering Problem

This paper addresses the problem of team orientation, a problem of the NP-hard class problems. To solve the problem, the metaheuristic Iterated Local Search (ILS) with Variable Neighborhood Descent (VND) was used. In ILS-VND, the choice of the sequence of neighborhood structures to be explored directly impacts the outcome. Because of this, Random VND and Multi-Armed Bandit were chosen to make better decisions regarding the exploration order. The objective of this study is to verify whether these variations with VND are good alternatives. Among the results obtained, with the choice of bad initial sequence, both the Bandit and the RVND were superior to the results of the VND. Compared to RVND, Bandit had a higher median in 63% of the cases tested, and in 56% of the cases, its worst results were higher than the worst results obtained with RVND.