Reverse Strategy for Non-Dominated Archiving

In the field of evolutionary multi-objective optimization (EMO), most EMO algorithms try to find a set of non-dominated solutions to approximate the Pareto front of a multi-objective optimization problem. In these algorithms, a population is evolved from one generation to another, and the population of the last generation is presented as the final result. However, recent studies reveal that some good solutions can be discarded during the evolutionary process, whereas these solutions are non-dominated. One way to solve this issue is to store all non-dominated solutions in an unbounded external archive (UEA) during the evolutionary process and select a set of solutions from the UEA as the final result. A recently proposed ND-Tree approach is very efficient for updating the UEA whenever a new solution is generated. However, this may not be the most efficient strategy. In this paper, we propose a simple yet efficient update strategy for the ND-Tree approach. The main idea is to reverse the order of solutions with respect to their generated time when updating the UEA. The experimental results suggest that the ND-Tree approach assisted by the proposed reverse strategy is much faster than the original ND-Tree approach in obtaining the final UEA. The optimal update frequency for the proposed strategy is also investigated.