The invention relates to an oscillation center voltage prediction method based on a polynomial model. The method includes the following steps that first, voltage value sampling and normalization calculation are performed, wherein in the actual running process of a power grid, the real-time voltage values of important transformer substations are acquired through a PMU, and normalization processing is performed on the voltage locus through the maximum value and the minimum value of oscillation recorded through offline simulation; second, the normalized voltage locus is predicted in real time, wherein the normalized voltage values in the first step and an observation window with a certain time length are utilized for curve fitting, the locus is fast predicted according to the polynomial model, and the normalized voltage values at future moments can be obtained; third, the actual predicted values are calculated, wherein counter-normalization processing is performed through the normalized voltage values obtained in the second step, the actual voltage values of the corresponding transformer substations at future moments are obtained, and then fast prediction of the oscillation center voltage locus is completed. Predicted values at future moments are provided for a transient unstability method based on a disturbed voltage locus through the oscillation center voltage prediction method, and guidance can be given to fast and effective discrimination of transient stability of the power grid.