关键词： 嵌入维数, 延迟时间, 相空间重构, BP神经网络, RBF神经网络

Abstract: The real-teime and procise short-ferm traffic flow forecesting is the key factor for the realizing of traffic control and traffic guidance in the intelligent traffic system.Saturated correlation dimension method and mutual information method are used to calculate embedding dimension and delay time,and the traffic flow time series is reconstructed accordingly in phase space.Wolf method is used to calculate the largest Lyapunov exponent,and the power spectrum of traffic flow time series is analyzed.Results show that the traffic flow series is a chaotic sequence with noise.The prediction models based on BP neural networks and RBF neural networks are applied to pedict traffic flow time series,which shows that the two models both  have good prediction effects,with the former having higher prediction accuracy and quicker prediction speed.

Key words: embedding dimension, delay time, phase space reconstruction, BP neural networks, RBF neural networks