Efficient constrained large-scale multi-objective optimization based on reference vector-guided evolutionary algorithm

The large-scale multi-objective optimization problem exist widely in reality while they have complex constraints. The simultaneous effect of the large-scale decision variables and the complexity of constraints makes the traditional multi-objective evolutionary algorithm face great challenges. For the large-scale of decision variables, some reference vector-guided, competitive group optimization-based and pairwise child generation-based algorithms have improved the search efficiency of constrained LSMOPs. However, these algorithms encounter difficulties in handling large-scale decision variables and complex constraints at the same time. In this paper, a reference vector-guided with dominance co-evolutionary multi-objective algorithm is proposed to solve constrained large-scale multi-objective problems. First, a reference vector is employed to guide several sub-populations with a fixed number of neighborhood solutions. Then, a new environmental selection is constructed using the angle penalty distance with dominance relationship. This new environmental selection strategy greatly enhances selection pressure. At the same time, a co-evolutionary constraint handling technology is applied to efficiently span the infeasible region. The proposed algorithm is evaluated on constrained large-scale multi-objective problems with 100, 500 and 1000 decision variables. In addition, the impact of each component of the proposed algorithm is examined for the overall performance of the algorithm and tested in a practical application in microgrids. The experimental results demonstrate the effectiveness of the algorithm in constrained large-scale multi-objective optimization.