Short-term prediction of traffic dynamics with real-time recurrent learning algorithms

Short-term prediction of dynamic traffic states remains critical in the field of advanced traffic management systems and related areas. In this article, a novel real-time recurrent learning (RTRL) algorithm is proposed to address the above issue. We dabble in comparing pair predictability of linear method versus RTRL algorithms and simple non-linear method versus RTRL algorithms individually using a first-order autoregressive time-series AR(1) and a deterministic function. A field study tested with flow, speed and occupancy series data collected directly from dual-loop detectors on a freeway is conducted. The numerical results reveal that the performance of RTRL algorithms in predicting short-term traffic dynamics is satisfactorily accepted. Furthermore, it is found that the dynamics of short-term traffic states characterised in different time intervals, collected in diverse time lags and times of day may have significant effects on the prediction accuracy of the proposed algorithms.