Coevolutionary self-tuning optimization algorithm

Abstract

The article proposes a coevolutionary self-tuning algorithm for solving global optimization problems. The algorithm simulates the selfish behavior of individuals of a herd of herbivores that are attacked by a pack of predators. Search agents are controlled by a set of attractive search operators based on the patterns of individual and collective behavior of agents, as well as the mechanisms of population selection in the «victim–predator» system. Agents move in the space of solutions to the optimization problem, using a set of operators that simulate various types of behavior, including selfish behavior. In contrast to most competing algorithms, the proposed coevolutionary self-tuning algorithm allows not only to model various types of selfish behavior. It includes computational mechanisms to maintain a balance between the convergence rate of the algorithm and the diversification of the solution search space. The effectiveness of the algorithm is analyzed using a series of experiments for the problems of finding the global minimum in a set of 5 known test functions. The results were compared with 7 competing bio heuristics by such indicators as the average best solution at the moment, the median best solution at the moment, and the standard deviation from the best solution at the moment. The accuracy of the proposed algorithm was higher than that of competing algorithms. A nonparametric proof of the statistical significance of the results obtained using the Wilcoxon T-test allows us to state that the results of the coevolutionary self-tuning algorithm are statistically significant.