A Multivariate Chaotic Time Series Approach for Road Network Short-Term Traffic State Forecasting

It is known that short-term traffic state forecasting is one of the most critical aspects in intelligent transportation systems (ITS). Previous attempts to forecast short-term traffic state concentrate on single-spot forecasting. Multi-spot forecasting based on multivariate chaotic time series analysis is proposed in this paper, where traffic states in different spots are considered as a whole. Multivariate time series derived from multi-spot traffic state data are reconstructed with time delays and embedding dimensions based on multivariate phase space reconstruction theory. Then performing forecasting model, multi-spot traffic state can be obtained from new input data. To verify that the proposed method performs better than univariate ones, real time data each 6 mins traffic volume of six cross-sections in six continuous spots on Beijing Second Loop-line expressway are illustrated.