Forecasting Traffic Volume at a Designated Cross-Section Location on a Freeway From Large-Regional Toll Collection Data

Abstract

Both road users and administrators are keen to know the traffic volume at the arbitrary point on the road network. In China, charging systems have been fully established in closed large-regional freeway networks. They have accumulated massive amounts of toll collection data and provided a possible method to forecast unknown traffic volume at any designated cross-section located on a freeway. A systematic method is proposed to derive the traffic volume step-by-step. First, the average traveling speed is obtained for each vehicle on its shortest path. Then, the traveling time is estimated in each road segment. Finally, the historical traffic volume is derived at the designated cross-section. To make the obtained traffic volume data more practical, a deep learning-based autoencoder is used for forecasting the traffic volume and evaluating its prediction accuracy. All these proposed methods are evaluated with a collection of toll data for one month covering more than 5000 km of freeway under a centralized regional charging system. One location is randomly selected as the designated cross section at 2 km from the upstream toll gate on a road segment of the Xi’an ring. The experimental results show the effectiveness and satisfactory accuracy of predicting the traffic volume in the designated cross-section compared with the data captured by the traffic video detection equipment. Rapid and successful prediction from available toll collection data may provide a practical method for deriving the traffic information without installing any additional regularly maintained detectors and equipment on the freeway.