A new decomposition-based evolutionary framework for many-objective optimization

A new class of Multi-Objective Evolutionary Algorithms (MOEAs) has emerged recently that uses the concept of decomposition to overcome the challenges faced by the current state-of-the-art MOEAs in undertaking optimization problems with more than three objectives. This new class of MOEAs employs a set of reference points to decompose the objective space into multiple scalar problems and to generate the target reference vectors for the solutions to sustain their diversity at every stage of the evolutionary process. In this study, we propose a novel framework for this class of MOEAs with a restricted mating selection scheme, with the aim to further improve the quality of the solutions close to the target reference vectors. The proposed framework is evaluated and compared with the current popular reference vector-based MOEAs to demonstrate its effectiveness. Using the Inverted Generation Distance (IGD) as the quality indicator, the experimental results indicate the superiority of the proposed framework when it is coupled with the MOEAs in solving 3- to 10-continuous objective functions in many-objective optimization problems.