Decomposition-based co-evolutionary algorithm for interactive multiple objective optimization

We propose a novel co-evolutionary algorithm for interactive multiple objective optimization, named CIEMO/D. It aims at finding a region in the Pareto front that is highly relevant to the Decision Maker (DM). For this reason, CIEMO/D asks the DM, at regular intervals, to compare pairs of solutions from the current population and uses such preference information to bias the evolutionary search. Unlike the existing interactive evolutionary algorithms dealing with just a single population, CIEMO/D co-evolves a pool of subpopulations in a steady-state decomposition-based evolutionary framework. The evolution of each subpopulation is driven by the use of a different preference model. In this way, the algorithm explores various regions in the objective space, thus increasing the chances of finding DM's most preferred solution. To improve the pace of the evolutionary search, CIEMO/D allows for the migration of solutions between different subpopulations. It also dynamically alters the subpopulations' size based on compatibility between the incorporated preference models and the decision examples supplied by the DM. The extensive experimental evaluation reveals that CIEMO/D can successfully adjust to different DM's decision policies. We also compare CIEMO/D with selected state-of-the-art interactive evolutionary hybrids that make use of the DM's pairwise comparisons, demonstrating its high competitiveness. (C) 2020 Elsevier Inc. All rights reserved.