An evolutionary many-objective algorithm based on decomposition and hierarchical clustering selection

• DOI: 10.1007/s10489-021-02669-9
• Publication Date: 2021-10-30T15:09:48Z
• Authors: Yuehong Sun; Kelian Xiao; Siqiong Wang; Qiuyue Lv
• Keywords: Rank (graph theory); Hierarchical clustering

In recent years, many multi-objective evolutionary algorithms have been proposed to solve many-objective optimization problems with regular Pareto front. These algorithms have shown good performance in balancing convergence and diversity. However, in the high-dimensional objective space, the non-dominated solutions increases exponentially as the number of objectives increases. The metrics to evaluate algorithm performance are also computationally intensive. In particular, solving the many-objective optimization problem of the irregular Pareto front faces great challenges. Moreover, many-objective evolutionary algorithms, do not easily show their convergence and diversity through visualization, as multi-objective evolutionary algorithms do. To address these problems, a many-objective optimization algorithm based on decomposition and hierarchical clustering selection is proposed in this paper. First, a set of uniformly distributed reference vectors divides non-dominanted individuals into different sub-populations, and then candidate solutions are selected based on the aggregation function values in the sub-populations. Second, a set of adaptive reference vectors is used to rank the dominant individuals in the population and retain promising candidate solutions. Third, a hierarchical clustering selection strategy is used to enable solutions with good convergence to be selected. Finally, a diversity maintenance strategy is used to remove solutions with poor diversity. The experimental results show that the proposed algorithm EA-DAH has advantages over other comparative algorithms in many-objective optimization problems with irregular Pareto fronts.