Traffic Flow Forecasting using the DESFM Model

Abstract

Acquiring the real-time information about traffic flow is one of the important steps toward the realization of ITS. For this purpose, we propose an approach for forecasting traffic flow. The proposed approach uses a double exponential smoothing (DES) model, first, to predict the future datum. Then, a Fourier series is employed to fit a residual series between the predicted series produced by the DES model and the actual observed series for all previous time steps, and to give a estimated value at the next step so as to correct the prediction of DES model. Furthermore, a Markov model is adopted to describe the transition behavior of the residual series and to construct a forecast trend adjustment scheme. Smoothing parameters of the DES model is determined by using Levenberg-Marquardt algorithm. The model integrated the Fourier residual correction scheme with the Markov forecast trend adjustment scheme is called DESFM model. The DESFM model is applied to predict the traffic flow of Route 23 in Nagoya-shi, Aichi-ken, and the performed results is compared with ARIMA prediction approach.