Nonlinear Multivariate Time-Space Threshold Vector Error Correction Model for Short Term Traffic State Prediction

The time-space threshold vector error correction (TS-TVEC) model is proposed and developed in this research for short term (hourly) traffic state prediction. The theory and method of cointegration with error correction mechanism is employed in the general design of the new statistical model TS-TVEC. An inherent connection is revealed between the error correction model and the transformed fundamental diagrams in macroscopic traffic flow theory. Error correction model is a linear model established on difference space, whereas, the fundamental relations between traffic variables exhibit piecewise linearity when they are transformed to difference space. Meanwhile, dynamics of traffic variables can be reflected in difference space. A threshold regime switching framework is implemented to deal with unknown structural breaks in non-stationary traffic time series to capture multiple traffic states. Spatial cross-correlated information is incorporated with a piecewise linear vector error correction model. As the TS-TVEC model is designed for nonlinear multivariate forecasting, it is compared to Neural Network, and Support Vector Regression model that are commonly used in this regard to test the effectiveness and robustness of the new statistical model. The TS-TVEC model was experimented on a small scale case in our empirical study and applied on a larger scale case thereafter. Both cases show that the TS-TVEC model is an effective tool that is capable of modelling the complexity of stochastic traffic processes and potentially applicable to real time traffic state prediction.