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.