A decomposition based evolutionary algorithm with direction vector adaption and selection enhancement

To further improve the search ability of the decomposition based many/multi-objective evolutionary algorithm (MOEA/D) in the tackling many-objective optimization problems (MaOPs) possessing complex characteristics (e.g., disconnected, degenerate, inverted, extremely convex or differently-scaled), we suggest an adaptive MOEA/D with better versatility, where the weight vector adaption and selection mechanism are improved. Firstly, a new niche-guided scheme by considering both the vector angle and Euclidean distance is proposed to leverage the search direction adaption upon different evolution phases, which is expected to be more robust for handling different types of irregular Pareto fronts (PFs). Secondly, in mating selection, a coordinated selection scheme aided by a multi-criterion decision procedure is utilized to enhance the effectiveness of recombination. Finally, in environmental selection, a steady state replacement strategy considering both the ensemble ranking of favorite subproblems with respect to solutions and improvement region restriction of subproblems is employed to alleviate misleading selection. Comparison experiments on benchmark MaOPs with diverse characteristics have been performed and the empirical results demonstrate the superiority of our proposal. The effects of direction vector adaption mechanism and other pertinent enhancements are also investigated.