Improving NSGA-III algorithms with information feedback models for large-scale many-objective optimization

Recently, more and more multi/many-objective algorithms have been proposed. However, most evolutionary algorithms only focus on solving small-scale multi/many-objective optimization problems, and few researchers pay attention to solving large-scale optimization problems. Like NSGA-III, although it performs well in solving many-objective optimization problems, it does not perform well in large-scale optimization problems. To solve this problem, in this paper, we introduce the information feedback models to improve the ability of NSGA-III to solve large-scale optimization problems. In the feedback models, the historical information of individuals in the previous generations is used in the update process of the current generation. According to the different ways of selecting individual, information feedback models are divided into two categories: one is to select individuals in a fixed way (called M-F) and the other is to select individuals randomly (called M-R). For each type of model, if the number of individuals selected is different, the information feedback models also have some difference. According to the number and mode of individual selection, there are six models in this paper. These six models are respectively embedded into NSGA-III to generate six improving NSGA-III algorithms (these six algorithms are collectively referred to as IFM-NSGAIII). These six algorithms are compared with the original NSGA-III on 9 benchmark problems to search the best information feedback model and the best IFM-NSGAIII algorithm. Then the two best IFM-NSGAIII algorithms are compared with four state-of-the-art algorithms on 9 test functions. Experiments show that the proposed algorithms are highly competitive on test problems.