Solving combinatorial problems with continuous swarm intelligence algorithms using machine learning techniques to select binarization schemes

Abstract

Today, combinatorial problems in binary domains are more frequently encountered in industry, so solving them efficiently is a priority in both academic and industrial areas. To solve the biggest problems the Metaheuristics have stood out in the last time. Some Metaheuristics have versions that make them capable of operating in discrete search spaces. But in the case of continuous swarm intelligence Metaheuristics, it is necessary to adapt them to operate in discrete domains. To make this adaptation it is necessary to use a binarization scheme, so as to take advantage of the original movements of the Metaheuristics designed for continuous spaces. In this work we propose a selector of binarization schemes based on Machine Learning techniques, selecting based on Q-Learning/SARSA the binarization to be used in each iteration, observing the balance of exploration and exploitation. To demonstrate the performance of the proposal, different forms of reward will be evaluated, which will be checked in an exhaustive way through the respective statistical tests, all this implemented on four Metaheuristics of continuous intelligence of swarms which are Sine Cosine Algorithm, Harris Hawk Optimization, Whale Optimization Algorithm and Grey Wolf Optimizer. The different Metaheuristics of continuous swarm intelligence using classical binary schemes will be compared against their Q-Learning enhanced versions.