SPSR-DMOEA algorithm improves dynamic multiobjective optimization with skeleton prediction

Dynamic multi-objective optimization problems change over time, making it difficult for algorithms to track the shifting Pareto front. Existing approaches often rely on individual migration or independent special point sampling, which fail to exploit the structural relationships among representative solutions. To address this, researchers GuangXian Gan and MinRong Chen propose SPSR-DMOEA, a Special Point Skeleton Reconstruction based Dynamic Multi-Objective Evolutionary Algorithm. The method begins by extracting three types of anchor points—the centroid, knee points, and extreme points—from the current Pareto optimal solution set. It then adaptively predicts their positions in the next environment based on their movement velocities across consecutive time steps.

The algorithm constructs a prediction skeleton in the decision space by linking the centroid to all other anchor points and building a minimum spanning tree (MST) among non-centroid nodes. This skeleton captures the overall population structure. Individuals are allocated to each skeleton edge proportionally based on edge length, and candidate solutions are generated uniformly along each edge. To prevent premature convergence, random orthogonal perturbations expand the search region around the skeleton. Experimental results on the DF dynamic benchmark suite confirm that SPSR-DMOEA achieves strong dynamic tracking performance, consistently outperforming baseline methods. The work highlights the value of leveraging geometric relationships among special points to guide population evolution under environmental changes.