The current nonlinear models of bolted joints ignore the influence of the effective contact area of bolts and deviate significantly from the actual assembly features of joints. In this regard， a dynamic modeling method of bolted joints considering local contact features is proposed. An orthogonal and nonlinear virtual material model with a cylindrical shape is introduced to express the flexible characteristics of joint interfaces under external load. Through establishing a stress-strain analysis model of micro-virtual material， the equations of calculating physical parameters of virtual materials are derived based on the principle of minimum potential energy and the Takashi Yoshimura method. A geometric parameter prediction model of the contact area of bolted joints is established based on the deep neural network （DNN） so that the nonlinear relationship mapping between the contact state of bolted joints and the diameter of the contact characteristic area is realized. Finally， a dynamic model of bolted joints with actual assembly characteristics is developed. Taking steel plates connected with bolts under a free boundary state as an experimental object， the effectiveness and accuracy of the proposed method are proved by comparing the result from the finite element simulation model established by the abovementioned method and experiments. The model developed by the proposed method shows a significant improvement in the comprehensive evaluation index of root mean square error （RMSE） compared with the virtual material modeling method without considering local contact features.