A Subspace-Based Method for PSO Initialization

Particle swarm optimization (PSO) is known to suffer under the curse of dimensionality. This paper proposes a novel strategy of particle swarm initialization particularly for high dimensional problems. The initialization strategy encourages the swarm to focus on exploitation rather than exploration, thereby allowing it to find fairly good solutions, even in the face of high dimensionality and very large search spaces. The proposed initialization strategy is compared to a number of other initialization strategies on high dimensional problems. The proposed strategy performed considerably better than all the other initialization strategies for the higher dimensional problems. Reasons for the observed behaviour are also discussed.