A Decomposition-Based Dynamic Reference Vector Adaptive Evolution Algorithm

Many predefined and improperly adjusted reference vectors in decomposition-based algorithms are invalid on irregular Pareto fronts (PFs). A dynamic reference vector adjustment based on a decomposition-based multi-objective evolutionary algorithm (DDMOEA) is proposed in this paper to address this problem. The algorithm separates the multi-objective problem into several sub-problems and sorts them according to dominance and diversity. A dynamic reference vector adjustment strategy is applied to guide the search area and obtain a set of optimal solutions that reflect PFs. By dynamically adjusting the reference vector, useless or overly close reference vectors can be deleted, and new reference vectors reflecting PFs can be generated, which is more conducive to the search process of the population. The proposed DDMOEA can improve the partitioning of sub-spaces and the performance of many objective problems with irregular PFs while considering computational complexity. In addition, in many objective optimization problems, the neighborhood adaptive strategy used in the algorithm reduces selection pressure while ensuring diversity. Experimental results prove that this algorithm can overcome multi-objective optimization problems with various categories of PFs compared to several of the most relevant algorithms.