This paper addresses the power system model order reduction proposing an optimal formulation for identifying electromechanical modes present in active power flows. The reduced power system preserves the dynamic and physical characteristics of the studied system. The modal identification is achieved by the digital Taylor-Fourier transform, which decomposes an oscillating signal into monocomponents. Then, these ones are embedded into an objective function with the purpose of estimating the equivalent generators' parameters. Results demonstrate a reduction in the computational burden and the computing time associated with transient stability studies for the preserved system, retaining its steady-state and dynamic conditions.