Multivariate arterial travel time prediction using hierarchical subspace vector autoregressive model

Short-term travel time prediction is an important function of Intelligent Transportation Systems. While there have been a lot of literatures on the freeway travel time prediction problem, it’s still a challenge for short-term arterial travel time prediction. Comparing with freeway traffic, vehicle flow on arterial network is more interruptive in a complex network context. So it’s important to take both the traffic flow character and spatial correlation into consideration for arterial travel time prediction. In this paper, we adopt multivariate time series models for modeling and predicting the travel time of multiple spatially correlated arterial links simultaneously. We propose a new hierarchical autoregressive model, namely Hierarchical Subspace Vector Autoregressive (HSVAR) model. The HSVAR model defines the interruptive factors as independent time series and expresses their impact to spatially correlated links with a mixing matrix. By leveraging the signal separation techniques, we devise an efficient algorithm for model evaluation under a heuristic optimization framework. The effectiveness of the proposed method is demonstrated through experiments on real world traffic data. And the comparison result with VARMA, ARMA and VAR models is also presented.