In this paper

a 3-layer PSO-trained BP Artificial Neural Networks（ANN） model is proposed for the on-line prediction of Bandwidth on Demand（BoD） ASON services and the Particle Swarm Optimization（PSO） algorithm improved.Particles with poor fitness are imposed with disturbances at random and then the disturbance effect is evaluated.The improved particles are accepted and at the same time the degenerated ones are also accepted in a certain probability.The maximum of the three predicted data is considered as the future bandwidth distribution

thus reducing the blocking rate of the bursting service.The simulation results show that this prediction model is appropriate for the on-line prediction of bursting and diversified BoD services.