For the microscopic analysis of road traffic conflicts in confluence sections of highway construction area and to explore the application of traffic conflict technology in the field of unmanned driving

12 variables of microscopic traffic information were collected as training set. The Bayesian network model was established combined with prior knowledge

and the precision of the model was evaluated by using cross validation method. Results show that large vehicles

as the main body of the conflict

could lead to the increase of the probability and severity of traffic conflicts. The preceding and following vehicles of two conflict vehicles had certain impact on the collision between the two vehicles. Overtaking behavior increased the probability of traffic conflicts. The probability and severity of traffic conflicts were higher in ramp merging sections than in turning sections. The proposed Bayesian network traffic conflict model was used to evaluate driving strategies

and conflict avoidance measures were put forward

such as diverging large vehicles. In certain road sections

it is suggested to take control measures to prohibit overtaking

limit the length of a single lane

set up distance confirmation facilities

and so on.