Abstract
A particle swarm optimization (PSO) algorithm with master-slave structure is proposed to train fuzzy wavelet neural network which be used to predict short-time traffic flow. The PSO algorithm is formulated in a form of hierarchical structure. The global search is performed at the master level, while the local search is carried out at the slave level. Through the harmonizing mechanism between master and slave level, the algorithm can execute global exact search without relying on complex coding operators. The simulation results demonstrate the proposed model can improve prediction accuracy, compared with BP based training techniques