Permafrost degradation is currently receiving as much attention as global warming. Permafrost evolution is a particularly slow process, making it impossible to examine the entire process by observation. Based on the heat conduction theory, in this study, a nonlinear numerical model for the entire process of permafrost evolution was developed with consideration of phase transition. The numerical simulation model was used to retrospectively analyse the evolution process and degradation pattern of permafrost in Tangchama beach-land in the Source Area of the Yellow River (SAYR) from 1400 to 2100; this analysis was performed using ice core inversion results obtained from the Qinghai-Tibet Plateau since the Little Ice Age and various future climate change scenarios from the IPCC Fifth Assessment Report. The simulation results showed that: 1) since 1900, permafrost temperatures have substantially increased, exceeding the range of change between 1400 and 1900; under the RCP2.6, RCP6.0 and RCP8.5 scenarios, permafrost will degrade and disappear in the next 100 years, and the degradation rate will subsequently accelerate. 2) The permafrost table steadily goes downward at first, and the onset of fast permafrost degradation is signalled by the appearance of a residual thawed layer at a depth of 3–4 m. The permafrost layer then gradually thins out as the permafrost table rapidly descends and the permafrost base slowly ascends. The permafrost will fully deteriorate and vanish once the permafrost table and the permafrost base bounds overlap. 3) Due to permafrost degradation, the ground temperature curve can be divided into the following stages: uniform gradient stage, C-type stage, zero-gradient type stage and phase transition stage. The four stages of the “permafrost life cycle” are initial temperature warming, continuous temperature rising, residual thawed layer, and final permafrost disappearance.