Short-time traffic flow prediction using third-order Volterra filter with product-decoupled structure

A prediction model for short-time traffic flow series is proposed in this paper. At first, estimation of the largest Lyapunov exponent is implemented by applying small data sets method so as to validate that chaos exists in traffic flow series. Then, through properly choosing the delay time and the embedding dimension using mutual information and false nearest neighbor methods, respectively, phase space reconstruction for traffic flow series is performed. In succession, aiming at the problem that number of coefficients for Volterra filter exponentially increases with the order of the filter, a third-order Volterra filter with approximately product-decoupled structure is put forward to reducing computational complexity. And the coefficients of this filter are adaptively adjusted employing an improved nonlinear normalized least mean square (NNLMS) algorithm. Finally, experimental results show that the proposed technique can effectively predict traffic flow series and reduce the model complexity.