Tackling the Bi-Objective Quadratic Assignment Problem by Characterizing Different Memory Strategies in a Memetic Algorithm

Several multi-objective evolutionary strategies have been successfully used to solve computationally complex optimization problems. The design contemplates many challenges: implementation of operators, parameterization, balance between genetic and local search operators, among others. The concept of memory appears implicitly in evolutionary strategies, since they retain the best characteristics of individuals throughout the generations. This work addresses the Bi-Objective Quadratic Assignment Problem (bQAP), characterizing the type of memory used in different heuristics, in order to maximize the performance in terms of solution coverage. Therefore, several heuristics have been proposed considering different search strategies, crossover and local search operators. The results show that a combination of memory strategies applied by memetic algorithms, increases the performance metrics. Although it was not possible to identify a memory strategy that obtains the best performance for all instances, it was possible to establish which of them contribute in the search of solutions considering the instances size. © 2017 IEEE.