Adaptive search space scaling in self-adaptive genetic algorithms for high-dimensional global black-box optimization problems

This paper proposes a adaptive search-space scaling (AdSS) mechanism for genetic algorithms and studies its influence on mixed continuous–discrete optimisation. The method combines binary encoding, clustering in the continuous subspace, and an Individual Cluster Efficiency (ICE) based procedure for adaptive modification of the search domain, complemented by elitism and immigration. The same mechanism is integrated into a self-configuring genetic algorithm, forming the SelfCGA+AdSS variant, while preserving the internal self-adaptation of evolutionary operators. The algorithms are evaluated on the CEC 2017 benchmark suite for multiple dimensionalities. The results demonstrate that AdSS consistently improves solution quality and reduces performance variability across runs, particularly on multimodal and composition functions. A detailed analysis of the composition function f₂₁ at D=10, supported by t-SNE population visualisations, reveals oscillatory convergence behaviour caused by interactions between adaptive domain contraction, cluster filtering, and diversity loss. The study confirms the effectiveness of dynamic search space scaling and identifies key factors influencing its stability.