A random dynamic grouping based weight optimization framework for large-scale multi-objective optimization problems

For large-scale multi-objective problems (LSMOPs), it is necessary to get a good grouping strategy or another way to reduce dimensions because of “the curse of dimensions”. In this paper, a weighted optimization framework with random dynamic grouping is proposed for large-scale problems. A weight optimization framework utilizes a problem transformation scheme in which weights are chosen to be optimized instead of the decision variables in order to reduce the dimensionality of the search space. Random dynamic grouping is used to determine sizes of each group adaptively. And multi-objective particle swarm optimization with multiple search strategies (MMOPSO) is employed as an optimizer for both original variables and weight variables. The proposed algorithm is performed on 28 benchmark test problems with 1000 dimensions, and the experimental results show that it can get better performance than some the-state-of-art algorithms in fewer function evaluations. In addition, it can be extended to solve LSMOPs with 5000 dimensions.