Interval island model initialization for permutation-based problems

In the absence of a priori knowledge about global optima, initial populations in genetic algorithms (GAs) should at least be diversified, especially while dealing with large spaces. On the other hand, the use of parallel models for GAs helps to solve large instances. We will focus on the island model. In this paper we propose an island initialization technique for permutation-based problems. We exploit a virtual tree organisation commonly used in exact methods (Branch and Bound) to generate a fully disjoint and well distributed (over the search space) initial population in each island. This method can be used for all permutation-based problems (QAP, Flow-shop, Q3AP..). regardless of the number of permutations. Experiments are performed over Q3AP benchmarks using a $10$ island model. The results shows the efficiency of the proposed method especially for large instances.