A hybrid swarm optimizer for efficient parameter estimation

Abstract

This paper proposes a hybrid algorithm for parameter estimation - a population-based, stochastic, particle swarm optimizer to identify promising regions of search space that are further locally explored by a Levenburg-Marquardt optimizer. This hybrid method is able to find global optimum for six benchmark problems. It is sensitive to the swarm topology which defines information transfer between particles; however, the hypothesis (Kennedy et al., 2001) that a star topology is better for finding the optimum for problems with large number of optima is not supported by this study. It is also seen that in the absence of the local optimizer, particle swarm alone is not as effective. The proposed method is also demonstrated on an identical catalytic reactor model.