Smart Starts: Accelerating Convergence through Uncommon Region Exploration

TL;DR

This work tackles the critical role of initialization in evolutionary algorithms for high-dimensional, multimodal optimization. It introduces OBLESA, a hybrid initialization that first uses Opposition-Based Learning to diversify the population and then employs a simplified Empty-Space Search Algorithm to fill under-explored regions, aiming to accelerate convergence. The approach is evaluated on 24 COCO benchmark functions with two EAs (DE and EGWO) and a fixed initial population size, showing improved convergence rates and solution quality, particularly in higher dimensions. The findings suggest that explicit targeting of empty regions, in combination with opposite solutions, yields more robust exploration and faster optimization in complex search spaces, with future work focused on refinement strategies and parameter tuning to further boost performance.

Abstract

Initialization profoundly affects evolutionary algorithm (EA) efficacy by dictating search trajectories and convergence. This study introduces a hybrid initialization strategy combining empty-space search algorithm (ESA) and opposition-based learning (OBL). OBL initially generates a diverse population, subsequently augmented by ESA, which identifies under-explored regions. This synergy enhances population diversity, accelerates convergence, and improves EA performance on complex, high-dimensional optimization problems. Benchmark results demonstrate the proposed method's superiority in solution quality and convergence speed compared to conventional initialization techniques.