The accurate identification of the critical machines cluster is the critical feature for transient stability analysis. Two novel relative transient kinetic and potential energy function are deduced by traditional transient energy function method which is based on pair-wise generators. According to the aggregation of these relative energy at fault clearing moment and two time steps later, a two-stage clustering algorithm is utilized to identify the critical machines cluster within a short fault cleared time. The derivation of the method shows that the proposed relative energy function relies on a part of machines' terminal electric parameters rather than a complete system network topology parameters or fault location information. So the proposed approach can reduce the omission error of important machines in a candidate set. And the identification time is reduced by the simple calculation progress. The correctness and effectiveness of the model are verified by simulation results of New England 10 generators 39 buses system and IEEE 145 buses system.