Although many multi-objective evolutionary algorithms (MOEAs) have been proposed, many-objective optimization problems (MaOPs) with more than three objectives remain difficult because convergence and diversity are hard to balance in high-dimensional objective spaces. To address this issue, this paper proposes a Population Classification-Based Evolutionary Algorithm (PCEA). First, the population is divided into subpopulations by a mapping-based classification method so that a baseline level of diversity can be maintained throughout evolution. Then, phase-wise subpopulation management strategies are employed. In Phase I, the sum of objective values is used as a practical convergence indicator within each subpopulation to accelerate convergence in the early stage. In Phase II, convergence and diversity are managed separately by non-dominated sorting and a new reference-point-based proximity indicator, respectively, so that diversity can be enhanced without losing convergence in the late stage. Experiments on the DTLZ1–DTLZ4 and WFG benchmark suites show that PCEA is competitive across a range of MaOPs.
Chen et al. (Mon,) studied this question.