A Q-Learning Hyperheuristic Binarization Framework to Balance Exploration and Exploitation

Many Metaheuristics solve optimization problems in the continuous domain, so it is necessary to apply binarization schemes to solve binary problems, this selection that is not trivial since it impacts the heart of the search strategy: its ability to explore. This paper proposes a Hyperheuristic Binarization Framework based on a Machine Learning technique of Reinforcement Learning to select the appropriate binarization strategy, which is applied in a Low Level Metaheuristic. The proposed implementation is composed of a High Level Metaheuristic, Ant Colony Optimization, using Q-Learning replacing the pheromone trace component. In the Low Level Metaheuristic, we use a Grey Wolf Optimizer to solve the binary problem with binarization scheme fixed by ants. This framework allowing a better balance between exploration and exploitation, and can be applied selecting others low level components.