The indoor thermal plume experiments and computational fluid dynamics（CFD）numerical simulations were used to analyze the axial velocity at different thermal plume intensities and compare the effects of different turbulence models on the simulation results. High-frequency sampling velocities in experiments were processed by wavelet transform. The characteristics of thermal plume were reflected in the mean velocity variation on low frequency information and turbulent pulsation on high frequency information. The results demonstrated the unsteady quasi-periodic fluctuation characteristics of thermal plume. The higher the surface temperature of the object, the larger the axial average velocity and turbulent pulsation amplitude. The simulation results of the unsteady Reynolds-averaged Navier-Stokes equations method eventually tend to be stable and fail to reflect the fluctuation characteristics of the thermal plume, while the large eddy simulation method can reflect the unsteady characteristics of the thermal plume. In addition, the implicit format time step can be set by the CFL (Courant-Friedrichs-Lewy) condition to obtain accurate simulation results.