Abstract: The laminar flow separation and separation-induced transition at low Reynolds number are complex, and have great difficulty in numerical simulation. Based on fully-connected back-propagation neural network, a data-driven model of intermittency at low Reynolds number was established. The input parameters of the data-driven model to reflect transition process and predict intermittency were selected through optimization design. By modifying the k-ω SST two equation turbulence model with a data-driven transition equation, the flow field evolution and unsteady aerodynamic characteristics of a two-dimensional airfoil under dynamic stall were solved. Results show that the data-driven transition equation combined with two equation turbulence model has the generalization ability for the angle of attack, and clearly reflects the typical flow conditions such as the growth and shedding of the leading-edge vortex and the reattachment of the flow under dynamic stall. The relative error of unsteady aerodynamic lift in dynamic stall between the data-driven transition model and the SST-γ three equation model is lower than 12%.