Decomposition-Based Interactive Evolutionary Algorithm for Multiple Objective Optimization

We propose a decomposition-based interactive evolutionary algorithm (EA) for multiple objective optimization. During an evolutionary search, a decision maker (DM) is asked to compare pairwise solutions from the current population. Using the Monte Carlo simulation, the proposed algorithm generates from a uniform distribution a set of instances of the preference model compatible with such an indirect preference information. These instances are incorporated as the search directions with the aim of systematically converging a population toward the DMs most preferred region of the Pareto front. The experimental comparison proves that the proposed decomposition-based method outperforms the state-of-the-art interactive counterparts of the dominance-based EAs. We also show that the quality of constructed solutions is highly affected by the form of the incorporated preference model.