A New Metaheuristic-Algorithm Similarity Measure Using Signal Flow Diagrams

Abstract

The component-based view for metaheuristic research promotes the identification of structural components of metaheuristics and metaheuristic-algorithms for analysis. In this study, we propose a method for measuring similarity between metaheuristic-algorithms. The method is based on a modified version of a signal flow representation of metaheuristic-algorithms that is aligned with the component-based view. The method takes any two metaheuristic-algorithms and decomposes them into their heuristic components whilst taking note of the order of execution of the heuristics. Features of the heuristics are then extracted, and finally a feature-based similarity calculation, that also considers the position of the heuristics, is performed to obtain an overall similarity score between the two metaheuristic-algorithms. The method incorporates more structural information in the similarity calculation than previous component-wise similarity measures and can be extended to cover a comprehensive set of metaheuristic components.