Differential Evolution With Preferential Interaction Network

Population-based metaheuristic optimization methods are built upon an algorithmic implementation of different types of complex dynamic behaviours. The problem-solving strategies they implement are often inspired by various natural and social phenomena whose fundamental principles were adopted for the use in practical search and optimization problems. New insights into complex systems, attained among others within the fields of network science and social network analysis, can be successfully incorporated into the study of evolutionary and swarm methods and used to improve their efficiency. Preferential attachment is a principle governing the growth of many real-world networks. That makes it a natural candidate for the use with network-based models of artificial evolution. Differential evolution is a widely-used evolutionary algorithm valued for its efficiency and versatility as well as simplicity and ease of implementation. In this paper, a variant of differential evolution, guided by an auxiliary model of population dynamics built with the help of the preferential attachment principle, is designed. The efficiency of the proposed approach is analyzed on the CEC 2017 real-parameter optimization benchmark.