Fuzzy Adaptive NSGA-III for Large-Scale Optimization Problems

doi:10.1007/s40815-021-01220-9

2022-01-16

PRE

Shanli Zhang

谢加良 (Jialiang Xie) *

Honghui Wang

DOI：10.1007/s40815-021-01220-9

原文链接

原文求助

摘要

En 中文

More and more multi-objective evolutionary algorithms are proposed and used to solve many-objective optimization problems and large-scale optimization problems. However, most of the existing algorithms use fixed crossover probability (pc) and mutation probability (pm) in the generation process of offspring, which makes the performance of the algorithm poor when dealing with complex problems. In this paper, by analyzing the complex non-linear relationship between performance metrics and pc and pm in the search process of many-objective evolutionary algorithms. A fuzzy inference system is constructed to dynamically update the pc and pm in the iterative process. Therefore, a fuzzy adaptive NSGA-III algorithm is proposed and used to solve large-scale optimization problems. For the construction of fuzzy systems, this paper takes the number of iterations, convergence metric, and diversity metric as inputs, and pc and pm as outputs. Four different fuzzy systems are obtained, and the best fuzzy system is selected through experiments. In order to further verify the effectiveness of the algorithm, the proposed algorithm and the existing literature are tested on LSMOP problems. The results show that the fuzzy system can well describe the complex non-linear relationship between the performance metrics and the pc and the pm in the search process of the many-objective evolutionary algorithm. It also effectively improves the performance of the algorithm when solving large-scale optimization problems, resulting in maintenance of the convergence and diversity of the population.

Keyword:

Fuzzy inference

Performance metrics

Crossover probability

Mutation probability

Large-scale optimization problems