Diversity Maintenance Mechanism for Multi-Objective Genetic Algorithms Using Clustering and Network Inversion

One of the major issues in applying multi-objective genetic algorithms to real-world problems is how to reduce the large number of evaluations. The simplest approach is a search with a small population size. However, the diversity of solutions is often lost with such a search. To overcome this difficulty, this paper proposes a diversity maintenance mechanism using clustering and Network Inversion that is capable of preserving diversity by relocating solutions. In addition, the proposed mechanism adopts clustering of training data sets to improve the accuracy of relocation. The results of numerical experiments on test functions and diesel engine emission and fuel economy problems showed that the proposed mechanism provided solutions with high diversity even when the search was performed with a small number of solutions.