MeEvo: Metacognitive Evolution Combined with Natural Evolution for Automatic Heuristic Design

MeEvo is a novel dual-layer framework for Automatic Heuristic Design (AHD) that integrates Natural Evolution and Metacognitive Evolution to enhance heuristic generation and refinement. By cyclically coupling these paradigms, MeEvo retains reasoning traces and fitness data while allowing for population-level recombination, leading to improved search efficiency and solution quality on complex optimization problems. Experimental results demonstrate that MeEvo outperforms existing LLM-based AHD architectures, particularly in handling constrained tasks, making it a significant advancement for practitioners in AI heuristic design.