Dynamic Procedure for Short-Term Prediction of Traffic Conditions

Abstract

Many existing models for forecasting traffic conditions are based on traffic flows. Field data are used here to show that these traffic conditions may not fluctuate from day to day in the same manner as does the traffic flow. Consequently, flow data are inappropriate for predicting traffic conditions because the same flow level may correspond to either a congested or a free-flow traffic state, a phenomenon that can be easily explained with the flow–density relationship. Occupancy, which is proportional to density, is a better indicator of traffic condition. A simple dynamic model based on occupancy data is proposed. The model utilizes occupancy and occupancy increments in an integrated way and treats them as two random variables represented by two normal distribution functions. It is shown that flow data, which are more stable than occupancy data, can be used indirectly to improve the performance of the proposed model. Self- and cross-validation efforts are made to examine the performance of the model. The results are promising. The expected absolute deviance for predicted occupancy (ranging from 0 to 100%) is about 1.25%, which is accurate enough for most applications. The model requires little effort in calibration and computation and is exceedingly simple to implement in the field.