Research Area: Big Data
Short-term traffic flow forecasting is a key problem in the area of intelligent transportation systems (ITS). Timely and accurate traffic state prediction is also the prerequisite of realizing proactive traffic control and dynamic traffic assignment effectively. In this paper, a new hybrid model for short-term traffic flow forecasting, which is built based on multifractal characteristics of traffic flow time series, is proposed. The hybrid model decomposes traffic flow series into four different components, namely a periodic part, a trend part, a stationary part and a volatility part, to unearth the traffic features hidden behind the data. Four parts are treated and modeled separately by using different methods, such as spectral analysis, time series and statistical volatility analysis, to further explore the underlying traffic patterns and improve forecasting accuracy. Performance of the proposed hybrid model is investigated with traffic flow data from freeway I-694 EB in the Twin Cities. The experimental results indicate that the proposed model outperforms in capturing nonlinear volatility and improving forecasting accuracy than traditional forecasting methods, especially for the multi-step ahead forecasting. Compared with the ARIMA-GARCH model, it gets an improvement of 8.23% in RMSE for one-step ahead forecasting and 10.69% for ten-step ahead forecasting. It is better than the hybrid model newly proposed in literature (Zhang et al. Transp Res Part C: Emerg Technol 43(1):65–78 2014) and gets an improvement of 1.27% in forecasting accuracy.
Keywords:
Author(s) Name: Hong Zhang, Xiaoming Wang, Jie Cao, Minan Tang and Yirong Guo
Journal name: Applied Intelligence
Conferrence name:
Publisher name: Springer
DOI: 10.1007/s10489-017-1095-9
Volume Information: volume 48, pages 2429–2440 (2018)
