The invention discloses a working condition on-line identification and early warning method of a low frequency oscillation leading model of an electric power system. A non-stationary signal is decomposed and handled by an ensemble empirical mode. By the adoption of a filter, a cross correlation coefficient and a signal energy weight coefficient, leading model components are screened out, a cross correlation function of the leading model components is obtained by a natural excitation technique, the cross correlation function is used as the leading model to identify signals, variable amplitudes and frequencies are identified by the adoption of a teager energy operator, phase positions are identified by the adoption of a time domain peak peak-value method, damping ratio is identified by the adoption of an energy analysis method and is applied to the working condition on-line early warning of low frequency oscillation leading model. The working condition on-line identification and early warning method has the advantages that the dynamic real-time leading model information of a system can be identified quickly and precisely and man-made excitation is not needed, and anti-noise performance is strong, and has an important meaning for maintaining a safe and stable on-line identification and early warning of the electric power system.