Decomposition-based evolutionary dynamic multiobjective optimization using a difference model

This paper presents a novel prediction model combined with a multiobjective evolutionary algorithm based on decomposition to solve dynamic multiobjective optimization problems. In our model, the motion of approximated Pareto-optimal solutions (POS) over time is represented by the motion of the centroid, and the other solutions are assumed to have the same motion as the centroid. A history of recent centroid locations is used to build a difference model to estimate the later motion of the centroid when an environmental change is detected, and then the new locations of the other solutions are predicted based on their current locations and the estimated motion. The predicted solutions, combined with some retained solutions, form a new population to explore the new environment, and are expected to track the new POS and/or Pareto-optimal front relatively well. The proposed algorithm is compared with four state-of-the-art dynamic multiobjective evolutionary algorithms through 20 benchmark problems with differing dynamic characteristics. The experimental studies show that the proposed algorithm is effective in dealing with dynamic problems and clearly outperforms the competitors.