Large Language Model-Aided Design of Hybrid COA, LSHADE and MPA Inspired Algorithm

Adopting large language models (LLMs) in metaheuristic optimization has opened up new possibilities for constructing highly sophisticated algorithms. This study offers a novel algorithm called the hybrid CLM, inspired by the crayfish optimization algorithm (COA), LSHADE, and marine predator algorithm (MPA), and designed using GPT-4o via the evolution of heuristics (EoH) prompting technique. The hybrid CLM adaptively switches between COA-inspired behavior stages for exploration, LSHADE-driven current-to-pbest mutation and crossover for exploitation, and MPA-style elite memory to guide population convergence through rank-aware recombination. The hybrid CLM is measured against five current state-of-the-art (SOTA) algorithms—COA, LSHADE, MPA, whale optimization algorithm (WOA), and sailfish optimization algorithm (SFO) across 64 functions (CEC2017, CEC2022, and mathematical benchmarks). Comparative analysis is performed using fitness, execution time, and convergence curve, supported by the Friedman rank and the Wilcoxon rank sum tests of significance. Results reveal that the hybrid CLM algorithm delivers both speed and accuracy, often outperforming or matching rivals in 24/29 (CEC2017), 7/12 (CEC2022), and 12/23 (mathematical). In CEC2017, the hybrid CLM’s execution time marked an overall 38% speed-up, six times faster than the MPA, with an insignificant difference in fitness performance with the MPA (rank difference = 1.29). Similar performance is also attained in CEC2022 and across the mathematical functions. Overall, the results demonstrate that this LLM-generated hybrid CLM can successfully compete with SOTA human-designed optimizers on difficult landscapes while cutting execution time by up to 40%. This successful proof of concept opens new possibilities for enhancing existing complex optimization algorithms by using LLMs as assistants while emphasizing their usefulness in designing metaheuristic algorithms.

Supervisor: Prof. Imtiaz Ahmad
Convener: Prof. Ayed Salman
Examination Committee: Dr. Mohammad Al-Failakawi.pdf