Metaheuristic Optimization Algorithms: A Comprehensive Survey of Taxonomy, Recent Advances, and Future Directions

Abstract

In recent decades, the metaheuristic optimization algorithms have emerged as an indispensable paradigm for tackling complicated non-convex, combinatorial, and complex problems that are not solved efficiently using classical gradient-based optimization algorithms. During the past three decades, hundreds of metaheuristics based on the concepts of evolution, collective behavior of animals, physical phenomena, human beings, etc., have been introduced. There has been extensive literature regarding different metaheuristic algorithms, although quite dispersed and fragmented at the same time. In this survey, we classify the metaheuristic optimization methods into five branches based on our taxonomy of evolutionary-based, swarm intelligence-based, physics-based, human-based, and hybrid or emerging methods, and review the principles, operators, advantages, and limitations of some of the landmark algorithms within each branch. We use a generic search loop structure to discuss the principle of balance between exploration and exploitation, which is the governing principle of the convergence behavior of all metaheuristic algorithms. Additionally, we consolidate some of the most popular benchmarks for constrained optimization and computational intelligence tasks in engineering applications, highlight performance results reported in the literature, and identify important open challenges, including scalability issues, sensitivity to tuning parameters, and premature convergence, among others. Finally, we discuss some of the emerging directions, including LLM-assisted meta-optimization, surrogate-assisted search for expensive objective functions, hybrid architectures, etc.