Proactive Assessment of Real time Traffic Flow Accident Risk on Urban Expressway

Abstract:

Based on dual loop detector data and accident data collected on Shanghai expressway, the Bayesian networks (BN) model was adopted for the modeling and analysis of real time traffic flow parameters and accident risk on expressways. Gaussian mixture model and the expectation maximization algorithm which could effectively deal with the missing data were also used in the parameters estimation of BN model. Then real time traffic safety risk was evaluated, and accident warning could be carried out in advance. Different combinations of dual loop detector data and time segments before accidents were used to develop the optimal accident risk estimation model by BN. The results show that the BN model adopting the nearest detector data upstream and downstream of the accident site within 5 to 10 minutes before the accident performs the best and the accident prediction accuracy is up to 76.94%. At last, a comparative study was made of the classical accident risk estimation algorithms including naive Bayes classifier, K nearest neighbor and back propagation (BP) neural network as well as the existing real time risk assessment studies. And the results show that the BN model obtains the best predictive results.