Full Elite Sets for Multi-Objective Optimisation

Copyright © 2002 Springer. The final publication is available at link.springer.com5th International Conference on Adaptive Computing in Design and Manufacture (ACDM 2002), Exeter, UK, 16-18 April, 2002Multi-objective evolutionary algorithms frequently use an archive of non-dominated solutions to approximate the Pareto front. We show that the truncation of this archive to a limited number of solutions can lead to oscillating and shrinking estimates of the Pareto front. New data structures to permit efficient query and update of the full archive are proposed, and the superior quality of frontal estimates found using the full archive is illustrated on test problems