Abstract: For particle swarm optimization (PSO), all particles are always initialized randomly, but here a new initialization method is presented to improve PSO optimizer convergence performance. On the basis of desired pattern, the corresponding aperture weights are solved by analytical techniques which can to a great extent ensure that these weights are effective estimations of the current optimum particle initial values. Then they are assigned to a particle as initial values, but all other particles of the swarm are still initialized randomly. Except this new initialization step, nothing is changed in PSO optimizer. The simulation results prove that this new optimizer converges faster and the fitness value converges deeper than the standard PSO especially in more complicated optimization problems. So the presented initialization method and new optimizer are effective in improving standard PSO convergence performances.

Key words: linear array, pattern optimization, nonlinear optimization, optimum particles, particle swarm optimization (PSO)