First results from a novel multispecies predator-prey optimization algorithm

Abstract

Optimization plays a central role in today's society. Different strategies have been proposed over the years, and their application is not independent of the problem. In this work, a novel predator-prey optimization algorithm (PPA) is presented and compared with genetic algorithms (GA). In the proposed algorithm, the search is based on the behaviour of a predefined number of species, some acting as predators and others as prey. The former pursue not only the latter to prey but also members of the same species to mate. On the other hand, preys run away from predators, but are attracted to other members of the same species. The performance of the proposed algorithm is tested with the help of four benchmark test functions: Goldstein-Price, Mishra's bird, Michalewicz and Eggholder functions. Each of these functions expose both PPA and GA to different difficulties. Five hundred runs were performed for each of the four test functions. Despite the better performance of the PPA compared to GA in all benchmarks analysed, the algorithm stands out when minimizing the Eggholder function, which has many local minima and the global minimum located near the edge of the search space. In the case of this function, the success rate is 75.6% for the PPA against 29.0% for the GA, when considering a tolerance of 1.0×10-2. The algorithm presented in this work has shown a resilient convergence to the global minimum with fewer iterations than the genetic algorithms, suggesting promising results in problems with similar characteristics.