Abstract
A radial basis function neural network (RBFNN) model is employed to predict the short-interval (within 15-minute) traffic series, including flow, speed and occupancy, which are measured in different time intervals, time lags, dimensions of state spaces, and times of day. Aside from describing entirely the methodology of RBFNN, the paper also uses two deterministic functions to test prediction power of the model. A field study with flow, time-mean-speed and percent occupancy time series directly extracted from two dual-loop detectors on a freeway of Taiwan is conducted. The results reveal that the predictive accuracies for different short-interval traffic dynamics by RBFNN model are quite satisfactory. It is also found that the predictive accuracies can be affected by the means of representing traffic series in terms of various time intervals, time lags, dimensions of state spaces, and times of day.
