Automated Design of Metaheuristics Using Reinforcement Learning within a Novel General Search Framework

Metaheuristic algorithms have been investigated intensively to address highly complex combinatorial optimisation problems. However, most metaheuristic designed manually by researchers of different expertise without a consistent framework support effective algorithm design. This paper proposes general search formulate in unified way range metaheuristics. defines generic algorithmic components, including selection heuristics and evolution operators. The aims serve as the basis analysing components for automated With established new framework, two reinforcement learning based methods, deep Q-network proximal policy developed automatically design population-based algorithm. proposed methods are able intelligently select combine appropriate during stages process. effectiveness generalization validated comprehensively across benchmark instances capacitated vehicle routing problem with time windows. study contributes making key step towards supporting fundamental analysis machine learning.