Research On Prediction Method Of Traffic Incident Duration

Prediction of traffic incident duration time is very important in Advanced Traffic Incident Management (ATIM). The duration of an incident is composed of four phases: detection time, response time, clearance time and recovery time. In the detection phase, the location, severity, and injuries of the incident are identified by the traffic managers, police, or patrol members. In the response phase, instrumented vehicles arrive at the incident site to handle the incident. In the clearance phase, the response team removes the obstacle that disturbs the traffic movements. In the recovery phase, queue starts to dissipate until the traffic restores to normal condition. The first three phases are called delay time. The purpose of this paper is to predict the duration of traffic incident, based on the incident data collected.Traffic incident data have been collected from different sources. One is from the Shanghai city expressway, the other is form highway in the Netherlands. Two kinds of prediction methods of traffic incident clearance time were presented in this research. The prediction method based on decision tree is applied to predict the traffic incident delay time in Shanghai. And the prediction model based on Bayesian decision tree is adopted to predict the clearance time in Netherlands. The Recovery time prediction model based on Cell Transmission Model (CTM) is presented, furthermore congestion delay and recovery delay models based on CTM are also developed.Firstly, this research design the traffic incident database of city expressway according to the current traffic incident management system in Shanghai, because the construction of the traffic incident database is the foundation of predicting traffic incident duration time. Several influence factors of incident duration time are analyzed in Shanghai city expressway, and prediction model of clearance time is developed based on decision tree method.Particular and accurate incident information has been collected by traffic incident management center in Netherlands, which is helpful for calibrating more precise prediction model of incident duration time. Therefore a new prediction model is presented based on the Bayesian decision tree. This model can deal with the information which is not complete or can not be obtained easily. Moreover the proposed model is demonstrated to be highly accurate, robust and reliable.The calibration method of the main parameters of the CTM model is studied, they are free flow speed, capacity of the bottleneck, the capacity, and the characteristic speeds. An example of the CTM model calibration is presented on the basis of a section of expressway in Shanghai. The results show that CTM model can simulate the real traffic situation and congestion characteristic actually when an incident occurs.Traffic incident recovery time and the congestion delay caused by the incident are calculated in this paper using CTM model. The influence factors of recovery time and traffic delay are analyzed, they are incident severity, incident clearance time, traffic demand. The results show that traffic recovery time accounts for large proportion in the total incident duration time in city expressway when traffic demand is high. The recovery delay also accounts for large proportion, and sometimes the recovery delay is even larger than the clearance delay. In this paper, the influence of many factors on the recovery time and recovery delay is studied quantitatively. The results can be used to optimize traffic incident management activities and improve the efficiency of city expressway.In the finality, future research directions are recommended.