A novel evolutionary framework based on a family concept for solving multi-objective bilevel optimization problems
Jesús-Adolfo Mejía-de-Dios, Alejandro Rodríguez-Molina, Efrén Mezura‐Montes · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2022
Multi-objective Bilevel Optimization (MOBO) problems are challenging because they include an optimization problem as part of the constraints within a multi-objective problem. Different evolutionary algorithms have been proposed to handle these MOBO problems, and they implement a solution conservative representation that was not designed for bilevel problems. This work presents a novel evolutionary framework based on a suitable and adaptive representation of MOBO problem solutions. The adopted representation is based on the family concept inspired by the biological classification of species, which helps to group representative solutions promoting diversity. Moreover, a non-dominated sorting and a density estimator (based on the hypervolume indicator) are adapted for the family concept. The proposed framework shows competitiveness in the experiments compared to a state-of-the-art algorithm for challenging constrained MOBO problems.