.png){kind=icon}
.png){kind=icon}
Return
A bi-layer decomposition algorithm for many-objective optimization problems
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-021-03135-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Decomposition-based evolutionary algorithms have shown great potential in solving many-objective optimization problems (MaOPs). However, manual parameters (e.g., neighbor size and scalarizing function) and the specified weight vector can easily degrade the performance of the algorithm, especially for MaOPs with irregular Pareto fronts. This paper presents a new decomposition-based evolutionary algorithm that adopts a bi-layer decision strategy to balance convergence and diversity of solutions for MaOPs, and it is free from neighborhood update and scalarizing methods. In the first layer decision, the well-converged solutions in each subregion form a primary population, where an adaptive fitness assignment considering the Pareto front shape is used to accelerate convergence. In the second layer decision, solutions in the primary population are ranked based on the diversity metric. The low-ranking solutions are added to the final population size until the population size is met. Further, to approximate the true PF as soon as possible, the intensity of convergence in the first layer decision and the activation frequency of the re-balance strategy are regulated. Moreover, we design a re-balance selection strategy to alleviate the dilemma of the specified weight vectors. The re-balance selection uses a clustering approach to adjust weight vectors to promote the uniform distribution of solutions. Finally, algorithms are verified on 150 test instances and one practical design problem. The experimental results show that the proposed algorithm performs better than five state-of-the-art peer algorithms on at least 64% test instances concerning hypervolume and has superiority over competitors on at least 72% test instances concerning the inverted generational distance plus.
Keywords:
Many-objective optimization
Adaptive decomposition
Bi-layer decision
Irregular Pareto front
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W