Short-term Traffic Status Prediction For Urban Expressway Based On Probe Vehicle Data

The prediction of urban road traffic status is a key to achieve ITS technologies such as future road traffic information querying platform and dynamic route guidance system.This paper studies the short-term traffic status prediction method for urban expressway, based on the consideration of the spatio-temporal relevance of traffic flow data and data missing condition. First, based on the summarization of traffic status prediction theory and the fact that probe vehicle data is relatively easy to be obtained in many China cities, link-based travel speed estimated from probe vehicle data is used as the short-term traffic status prediction indicator of expressway. Second, the missing data is complemented with a method which combined statistical method and K-means classification approach, and data de-noising based on wavelet transformation approach was applied to reduce error. Third, through the analysis of spatio-temporal relevance of travel speed on links, combined with the characteristics of the traffic flow data itself and the advantages and disadvantages of all kinds of traffic status prediction models, BP Neural Network Model based traffic status prediction models are established respectively using spatial, temporal and spatio-temporal feature vectors. Finally, using the data collected from an expressway road section long about3.2km in Beijing, the proposed models are evaluated.The results showed that all the models proposed can meet the requirement of short-term traffic status prediction accuracy. The model based on spatio-temporal feature vector has the best accuracy up to98.3%tested by links participated in model establishing and to93.64%tested by other links not participated in model establishing in the expressway section. The model based on temporal feature vector shows the second best with98.2%accuracy tested by links participated in model establishing and to92.84%tested by other links not participated in model establishing in the expressway section, while the model based on spatial feature vector shows the last with95%accuracy tested by links participated in model establishing and to91.94%tested by other links not participated in model establishing in the expressway section. Consequently, this paper proposes a comprehensive model for traffic status prediction based on data missing condition. In detail, when both time and space related data are available, the model based on spatio-temporal feature vector is applied; when time related data is not enough or unavailable, the model based on temporal feature vector is applied; and when space related data is not enough or unavailable, the model based on spatial feature vector is applied.