A Global Optimization Approach Based on Opinion Formation in Complex Networks

Population-based metaheuristic optimization techniques have numerous applications in science and engineering. In this paper, we introduce a novel population-based binary optimization method, built upon consensus formation in interacting multi-agent systems. Agents, each associated with an opinion vector, are linked together through a network structure. The agents influence each other by performing interactions, and as a result their opinions evolve. Opinion vectors hold solutions to the problem, and at the same time, store additional information on agents’ interaction experience. The agents communicate and work collectively to solve an optimization task. In this study, we consider a specific opinion update rule and various underlying interaction network topologies. Results of the experiments, conducted on a number of benchmark cost functions, show that a dynamical ring topology, designed for our specific purpose, leads to the best performance compared to other network topologies. We also compare the performance of the proposed optimization algorithm with classical and state-of-the-art population-based optimizers, namely, Genetic Algorithm, Binary Hybrid Topology Particle Swarm Optimization, Selectively Informed Particle Swarm Optimization, Binary Learning Differential Evolution, and Discrete Artificial Bee Colony. Comparisons based on experimental analysis reveal that the proposed consensus-based optimizer is the top-performer.