Preliminary study of adaptive grid-based decomposition on many-objective evolutionary optimization

This work proposes a method to adaptively determine the grid granularity for the grid decomposition-based evolutionary algorithm for many-objective optimization. The decomposition of the objective space is one of the promising approaches for evolutionary many-objective optimization. Although the weight-based decomposition is frequently employed recently, it has been known that the grid-based decomposition is also effective for many-objective optimization. The grid-based evolutionary algorithm (GrEA) decomposes the objective space into a grid environment with the grid division parameter. The grid division parameter determines the grid granularity and affects search performance. To find out an appropriate grid division parameter, we need to perform a parameter tuning, which executes multiple runs with different parameters and is generally time-consuming. In this work, we propose an adaptive GrEA (A-GrEA) automatically determines the grid division parameter during a single run of the search. Experimental results using WFG variant problems with 6 to 10 objectives show that the proposed A-GrEA does not only to determine the grid division parameter adaptively but also achieves comparative or higher search performance than the conventional GrEA with the best grid division parameter, MOEA/D, and NSGA-III.