Abstract:

In this paper, 496 traffic accidents in Qinling Tunnel group on Xi′anHanzhong Expressway are taken as study objects to analysis the road environmental factors affecting traffic accident prediction of highway tunnels; the main variables having significant influence on different prediction class are selected; the traffic accident form, accidental severity, accident casualty and accident duration of highway tunnel are predicted by naive Bayesian model, random forest model, BP neural network model and support vector machine model, respectively; and the optimal prediction model is determined based on accuracy and stability of each model. The study results show that: (1) The random forest model is most suitable for prediction of highway tunnel accident form, and its accuracy reaches about 84%. (2) The naive Bayesian model should be given priority to the prediction of traffic accident severity and casualty of highway tunnels, whose accuracy is as high as 50% when predicting major or extraordinary serious accidents. (3) When the random forest model is selected as the prediction model for highway tunnel traffic accident duration, the accuracy exceeds 70% when the absolute error is 20 min.

Key words: highway tunnel, traffic accident, prediction model, correlation analysis, highway environment