Short-term Traffic Forecasting Using Multivariate Autoregressive Models

Dmitry Pavlyuk

Procedia Engineering (1877-7058), Vol. 178, pp. 57-66(2017)
DOI: 10.1016/j.proeng.2017.01.062
Keywords: traffic flow; short-term forecasting; multivariate modelling; time series; spatio-temporal modelling

Open access: DOWNLOAD FULL TEXT  Export citation: BibTeX RIS


This research is devoted to a systematic review of multivariate models in the context of their application to short-term traffic flow forecasting. A set of discussed models includes autoregressive integrated moving average models (ARIMA and VARMA), error correction models (VECM and EC-VARMA), space-time ARMA (STARMA), and multivariate autoregressive space state models (MARSS). All these models are based on different assumptions about a structure of interrelationships in traffic data (in time, in space or between different traffic characteristics). We discussed base assumptions of these models (such as stationary of traffic flows and spatial independence of data) and their importance in the domain of transport flows. The discussion is supplemented with an empirical application of the models to forecasting of traffic speeds for a small road segment. Empirical conclusions and projected research directions are also presented.