A Preference-based Multiobjective Evolutionary Algorithm Based on Weight Vector Adjustment Strategy

Evolutionay multiobjective optimization (EMO) methodologies have gained popularity in finding well-distributed trade-off solutions approximating to the pareto-optimal front (PF) in the past decades. In preference-based multiobjective optimization, the decision maker (DM) only be interested in a partial region, called ROI. The DM can integrate preference information into optimization to guide the search for ROI and accelerate the convergence of the population. In this paper, we propose a decomposition-based interactive evolutionary algorithm for preference-based multiobjective optimization problems (MOEA/D-WVA). Differ from most existing decomposition-based algorithms, MOEA/D-WVA guides the population to converge to the ROI by transforming the originally evenly distributed reference vectors into a biased distribution. In addition, we propose an interactive preference model in this paper by which the DM can specifies preference information progressively during the search process. The proposed algorithm is compared with three state-of-the-art preference-based multiobjective evolutionary algorithms on a variety of benchmark problems. Experimental results demonstrate the effectiveness of our proposed algorithm for approximating the preferred solutions in the ROI.